Skip to content

Using TensorFlow to assess bank telemarketing campaign impact

License

Notifications You must be signed in to change notification settings

scriptbuzz/ml-bank-telemarketing

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

7 Commits
 
 
 
 
 
 

Repository files navigation

Bank Telemarketing Impact Analysis and Prediction

NAME: Michael H Bitar

DATE: Sep 5, 2022 / r0.3

EMAIL: seeds.swipes.0t@icloud.com

PROJECT: Develop a machine learning solution to assist marketing executives at a client bank to understand which characteristics of potential customers are the best predictors of purchasing of one of the bank’s products. The client bank is interested in using a predictive model to score each potential customer’s propensity to purchase, as well as understand which customer characteristics are most important in driving purchasing behavior, in order to inform future marketing segmentation personalization.

METHODOLOGY: The provided dataset is imbalanced. I will be using various under/over sampling methodology to search for the best F1 Score.

RESULTS: Oversampling the minority class provided the highest F1 Score for both classes in target y.

REFERENCES: Content of this guide is based on authors listed in the references section at the end of this notebook. Thanks to all the authors for their valuable educational contributions.

# Import needed packages

import numpy as np
import pandas as pd
import tensorflow as tf
import sklearn as sk

from sklearn.preprocessing import MinMaxScaler
from sklearn.model_selection import train_test_split
from sklearn.metrics import confusion_matrix , classification_report

import matplotlib.pyplot as plt
import seaborn as sns

from pandas.plotting import scatter_matrix
from pandas_profiling import ProfileReport

# set random seeds for reproducability 
tf.random.set_seed(100)
seed = 15

# disable warning messages when possible
import warnings
warnings.filterwarnings('ignore')

# display all dataframe columns
pd.set_option('display.max_columns', None)

# suppress sci notation for readability
np.set_printoptions(suppress=True)

# Make numpy values easier to read
np.set_printoptions(precision=3, suppress=True)

tf.keras.layers.Normalization 
from tensorflow.keras import layers
from tensorflow import keras
from tensorflow_addons import losses
# load dataset over the web

url = "https://raw.githubusercontent.com/scriptbuzz/datasets/main/bank-full.csv"
df = pd.read_csv(url,sep=";",error_bad_lines=False, engine ='python')
df.info()
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 45211 entries, 0 to 45210
Data columns (total 17 columns):
 #   Column     Non-Null Count  Dtype 
---  ------     --------------  ----- 
 0   age        45211 non-null  int64 
 1   job        45211 non-null  object
 2   marital    45211 non-null  object
 3   education  45211 non-null  object
 4   default    45211 non-null  object
 5   balance    45211 non-null  int64 
 6   housing    45211 non-null  object
 7   loan       45211 non-null  object
 8   contact    45211 non-null  object
 9   day        45211 non-null  int64 
 10  month      45211 non-null  object
 11  duration   45211 non-null  int64 
 12  campaign   45211 non-null  int64 
 13  pdays      45211 non-null  int64 
 14  previous   45211 non-null  int64 
 15  poutcome   45211 non-null  object
 16  y          45211 non-null  object
dtypes: int64(7), object(10)
memory usage: 5.9+ MB
# # Profile the dataset 
# profile = ProfileReport(df, title="Pandas Profiling Report")
# profile
# Generate scatter diagrams

scatter = scatter_matrix(df[['age','balance','duration','campaign']],figsize=(10,10))
plt.suptitle('Scatter Matrix of Age, Balance, Duration and Campaign')
plt.show()

png

# show any duplicates rows in dataframe
df.duplicated().sum()
0
# show columns and their  uique values 
def print_unique_col_values(df):
       for column in df:
            if df[column].dtypes=='object':
                print(f'{column}: {df[column].unique()}') 
print_unique_col_values(df)
job: ['management' 'technician' 'entrepreneur' 'blue-collar' 'unknown'
 'retired' 'admin.' 'services' 'self-employed' 'unemployed' 'housemaid'
 'student']
marital: ['married' 'single' 'divorced']
education: ['tertiary' 'secondary' 'unknown' 'primary']
default: ['no' 'yes']
housing: ['yes' 'no']
loan: ['no' 'yes']
contact: ['unknown' 'cellular' 'telephone']
month: ['may' 'jun' 'jul' 'aug' 'oct' 'nov' 'dec' 'jan' 'feb' 'mar' 'apr' 'sep']
poutcome: ['unknown' 'failure' 'other' 'success']
y: ['no' 'yes']
# check value for uniquness of target label
df['y'].unique()
array(['no', 'yes'], dtype=object)
# count each unique value for target label. 
# We can see an imbalance in favor of "no" values for the traget label

df.y.value_counts()
no     39922
yes     5289
Name: y, dtype: int64
# apply nominal one-hot encoding to select columns
df = pd.get_dummies(data=df, columns=['job','education','marital','month','poutcome','contact'])
df.columns
Index(['age', 'default', 'balance', 'housing', 'loan', 'day', 'duration',
       'campaign', 'pdays', 'previous', 'y', 'job_admin.', 'job_blue-collar',
       'job_entrepreneur', 'job_housemaid', 'job_management', 'job_retired',
       'job_self-employed', 'job_services', 'job_student', 'job_technician',
       'job_unemployed', 'job_unknown', 'education_primary',
       'education_secondary', 'education_tertiary', 'education_unknown',
       'marital_divorced', 'marital_married', 'marital_single', 'month_apr',
       'month_aug', 'month_dec', 'month_feb', 'month_jan', 'month_jul',
       'month_jun', 'month_mar', 'month_may', 'month_nov', 'month_oct',
       'month_sep', 'poutcome_failure', 'poutcome_other', 'poutcome_success',
       'poutcome_unknown', 'contact_cellular', 'contact_telephone',
       'contact_unknown'],
      dtype='object')
df.info()
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 45211 entries, 0 to 45210
Data columns (total 49 columns):
 #   Column               Non-Null Count  Dtype 
---  ------               --------------  ----- 
 0   age                  45211 non-null  int64 
 1   default              45211 non-null  object
 2   balance              45211 non-null  int64 
 3   housing              45211 non-null  object
 4   loan                 45211 non-null  object
 5   day                  45211 non-null  int64 
 6   duration             45211 non-null  int64 
 7   campaign             45211 non-null  int64 
 8   pdays                45211 non-null  int64 
 9   previous             45211 non-null  int64 
 10  y                    45211 non-null  object
 11  job_admin.           45211 non-null  uint8 
 12  job_blue-collar      45211 non-null  uint8 
 13  job_entrepreneur     45211 non-null  uint8 
 14  job_housemaid        45211 non-null  uint8 
 15  job_management       45211 non-null  uint8 
 16  job_retired          45211 non-null  uint8 
 17  job_self-employed    45211 non-null  uint8 
 18  job_services         45211 non-null  uint8 
 19  job_student          45211 non-null  uint8 
 20  job_technician       45211 non-null  uint8 
 21  job_unemployed       45211 non-null  uint8 
 22  job_unknown          45211 non-null  uint8 
 23  education_primary    45211 non-null  uint8 
 24  education_secondary  45211 non-null  uint8 
 25  education_tertiary   45211 non-null  uint8 
 26  education_unknown    45211 non-null  uint8 
 27  marital_divorced     45211 non-null  uint8 
 28  marital_married      45211 non-null  uint8 
 29  marital_single       45211 non-null  uint8 
 30  month_apr            45211 non-null  uint8 
 31  month_aug            45211 non-null  uint8 
 32  month_dec            45211 non-null  uint8 
 33  month_feb            45211 non-null  uint8 
 34  month_jan            45211 non-null  uint8 
 35  month_jul            45211 non-null  uint8 
 36  month_jun            45211 non-null  uint8 
 37  month_mar            45211 non-null  uint8 
 38  month_may            45211 non-null  uint8 
 39  month_nov            45211 non-null  uint8 
 40  month_oct            45211 non-null  uint8 
 41  month_sep            45211 non-null  uint8 
 42  poutcome_failure     45211 non-null  uint8 
 43  poutcome_other       45211 non-null  uint8 
 44  poutcome_success     45211 non-null  uint8 
 45  poutcome_unknown     45211 non-null  uint8 
 46  contact_cellular     45211 non-null  uint8 
 47  contact_telephone    45211 non-null  uint8 
 48  contact_unknown      45211 non-null  uint8 
dtypes: int64(7), object(4), uint8(38)
memory usage: 5.4+ MB
# apply binary encoding to select columns

df['housing'] = df['housing'].map({'yes': 1, 'no': 0})
df['loan'] = df['loan'].map({'yes': 1, 'no': 0})
df['default'] = df['default'].map({'yes': 1, 'no': 0})
df['y'] = df['y'].map({'yes': 1, 'no': 0})
df.sample(20)
age default balance housing loan day duration campaign pdays previous y job_admin. job_blue-collar job_entrepreneur job_housemaid job_management job_retired job_self-employed job_services job_student job_technician job_unemployed job_unknown education_primary education_secondary education_tertiary education_unknown marital_divorced marital_married marital_single month_apr month_aug month_dec month_feb month_jan month_jul month_jun month_mar month_may month_nov month_oct month_sep poutcome_failure poutcome_other poutcome_success poutcome_unknown contact_cellular contact_telephone contact_unknown
19134 38 0 202 0 0 5 1311 2 -1 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 0 1 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 0 0
2906 49 0 387 1 0 14 347 1 -1 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 1
6083 51 0 51 0 0 27 121 2 -1 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 1
22974 52 0 60 0 1 26 196 2 -1 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 0 0
21472 58 0 2180 0 0 18 273 4 -1 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 0 0
12657 27 0 35 0 0 4 255 1 -1 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 1 0 0
34582 39 0 782 1 0 5 542 4 47 11 0 1 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 0 1 0 0
10749 56 0 21664 0 0 17 214 8 -1 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0 1
41068 26 0 535 0 0 14 125 1 -1 0 1 0 1 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 0 0
5847 31 0 -452 1 0 26 182 5 -1 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 1 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 1
32866 35 0 520 1 0 17 476 1 -1 0 1 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 0 0
39582 34 0 23076 0 0 26 243 2 35 6 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 0 1 0 0
9398 58 0 1657 0 0 6 172 2 -1 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0 1
40101 27 0 538 0 0 4 124 2 -1 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 1 1 0 0
31400 24 0 1250 0 0 25 411 2 -1 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 1 1 0 0
7868 35 0 2708 1 0 30 178 2 -1 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 1
308 44 0 129 1 1 5 189 2 -1 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 1
18300 46 0 272 0 0 31 149 4 -1 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 1 0 0
5307 34 0 2144 1 0 23 157 1 -1 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 1
4807 35 0 -15 1 0 21 554 3 -1 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 1
# normalize select data

columns_to_scale = ['age','balance','day','duration','campaign','pdays','previous']
scaler = MinMaxScaler()
df[columns_to_scale] = scaler.fit_transform(df[columns_to_scale])
df
age default balance housing loan day duration campaign pdays previous y job_admin. job_blue-collar job_entrepreneur job_housemaid job_management job_retired job_self-employed job_services job_student job_technician job_unemployed job_unknown education_primary education_secondary education_tertiary education_unknown marital_divorced marital_married marital_single month_apr month_aug month_dec month_feb month_jan month_jul month_jun month_mar month_may month_nov month_oct month_sep poutcome_failure poutcome_other poutcome_success poutcome_unknown contact_cellular contact_telephone contact_unknown
0 0.519481 0 0.092259 1 0 0.133333 0.053070 0.000000 0.000000 0.000000 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 1
1 0.337662 0 0.073067 1 0 0.133333 0.030704 0.000000 0.000000 0.000000 0 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 1
2 0.194805 0 0.072822 1 1 0.133333 0.015453 0.000000 0.000000 0.000000 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 1
3 0.376623 0 0.086476 1 0 0.133333 0.018707 0.000000 0.000000 0.000000 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 1
4 0.194805 0 0.072812 0 0 0.133333 0.040260 0.000000 0.000000 0.000000 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 1
... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ...
45206 0.428571 0 0.080293 0 0 0.533333 0.198658 0.032258 0.000000 0.000000 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 1 1 0 0
45207 0.688312 0 0.088501 0 0 0.533333 0.092721 0.016129 0.000000 0.000000 1 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 1 1 0 0
45208 0.701299 0 0.124689 0 0 0.533333 0.229158 0.064516 0.212156 0.010909 1 0 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 1 0 0
45209 0.506494 0 0.078868 0 0 0.533333 0.103294 0.048387 0.000000 0.000000 0 0 1 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 1 0 1 0
45210 0.246753 0 0.099777 0 0 0.533333 0.073404 0.016129 0.216743 0.040000 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 1 0 0

45211 rows Ă— 49 columns

for col in df:
    print(f'{col}: {df[col].unique()}')
age: [0.519 0.338 0.195 0.377 0.221 0.13  0.312 0.325 0.299 0.143 0.455 0.506
 0.429 0.351 0.545 0.494 0.182 0.091 0.286 0.273 0.442 0.364 0.234 0.403
 0.532 0.247 0.416 0.468 0.481 0.39  0.078 0.26  0.169 0.156 0.117 0.208
 0.065 0.104 0.558 0.052 0.039 0.026 0.623 0.571 0.844 0.74  0.636 0.675
 0.61  0.649 0.597 0.662 0.701 0.688 0.013 0.753 0.87  0.584 0.935 0.831
 0.714 0.727 0.779 0.805 0.987 0.792 0.766 0.883 1.    0.818 0.    0.922
 0.857 0.896 0.961 0.974 0.909]
default: [0 1]
balance: [0.092 0.073 0.073 ... 0.147 0.202 0.221]
housing: [1 0]
loan: [0 1]
day: [0.133 0.167 0.2   0.233 0.267 0.367 0.4   0.433 0.467 0.5   0.6   0.633
 0.667 0.733 0.833 0.867 0.9   0.933 0.967 0.033 0.067 0.1   0.333 0.533
 0.567 0.767 0.8   0.    0.3   0.7   1.   ]
duration: [0.053 0.031 0.015 ... 0.264 0.253 0.316]
campaign: [0.    0.016 0.032 0.065 0.048 0.081 0.097 0.113 0.129 0.145 0.161 0.177
 0.194 0.29  0.21  0.371 0.242 0.5   0.274 0.339 0.226 0.258 0.387 0.323
 0.677 0.806 1.    0.645 0.403 0.435 0.871 0.79  0.597 0.355 0.306 0.452
 0.484 0.581 0.468 0.726 0.419 0.919 0.516 0.548 0.532 0.565 0.613 0.694]
pdays: [0.    0.174 0.192 0.106 0.1   0.165 0.17  0.103 0.162 0.203 0.117 0.201
 0.196 0.193 0.225 0.19  0.149 0.217 0.226 0.198 0.136 0.138 0.12  0.197
 0.135 0.189 0.153 0.151 0.142 0.183 0.214 0.128 0.133 0.134 0.2   0.205
 0.127 0.175 0.111 0.119 0.173 0.202 0.222 0.209 0.213 0.178 0.167 0.159
 0.146 0.208 0.126 0.182 0.194 0.112 0.21  0.147 0.15  0.224 0.144 0.122
 0.118 0.031 0.206 0.033 0.211 0.179 0.13  0.139 0.158 0.143 0.216 0.219
 0.131 0.187 0.155 0.195 0.218 0.01  0.166 0.22  0.212 0.204 0.007 0.115
 0.154 0.108 0.107 0.013 0.116 0.18  0.228 0.123 0.177 0.169 0.148 0.009
 0.14  0.185 0.124 0.104 0.032 0.227 0.157 0.161 0.141 0.181 0.172 0.156
 0.036 0.132 0.114 0.221 0.188 0.04  0.11  0.163 0.037 0.229 0.109 0.125
 0.034 0.308 0.284 0.291 0.26  0.281 0.275 0.282 0.235 0.266 0.274 0.297
 0.265 0.292 0.305 0.083 0.257 0.283 0.288 0.306 0.276 0.236 0.3   0.298
 0.278 0.299 0.269 0.289 0.259 0.186 0.273 0.302 0.286 0.294 0.253 0.261
 0.237 0.258 0.287 0.271 0.263 0.303 0.003 0.311 0.267 0.29  0.239 0.231
 0.31  0.268 0.295 0.314 0.313 0.279 0.304 0.24  0.247 0.256 0.312 0.234
 0.255 0.233 0.249 0.232 0.296 0.264 0.242 0.25  0.087 0.245 0.085 0.088
 0.307 0.243 0.248 0.089 0.272 0.095 0.008 0.241 0.315 0.002 0.28  0.244
 0.317 0.093 0.318 0.011 0.321 0.015 0.322 0.102 0.319 0.099 0.097 0.252
 0.029 0.025 0.325 0.048 0.338 0.057 0.378 0.353 0.349 0.381 0.354 0.345
 0.075 0.361 0.33  0.38  0.382 0.347 0.372 0.366 0.383 0.07  0.375 0.385
 0.36  0.359 0.351 0.374 0.376 0.386 0.356 0.377 0.37  0.046 0.364 0.336
 0.339 0.357 0.352 0.368 0.365 0.333 0.067 0.369 0.164 0.39  0.346 0.362
 0.388 0.384 0.391 0.367 0.021 0.086 0.171 0.392 0.344 0.396 0.393 0.373
 0.397 0.398 0.35  0.323 0.394 0.389 0.017 0.399 0.018 0.335 0.4   0.401
 0.328 0.403 0.327 0.03  0.326 0.32  0.094 0.006 0.101 0.096 0.092 0.081
 0.016 0.337 0.044 0.091 0.073 0.026 0.341 0.408 0.077 0.023 0.041 0.414
 0.411 0.407 0.404 0.416 0.412 0.42  0.343 0.329 0.419 0.417 0.055 0.415
 0.331 0.421 0.409 0.405 0.413 0.342 0.422 0.406 0.423 0.049 0.334 0.078
 0.427 0.425 0.424 0.058 0.042 0.429 0.43  0.428 0.358 0.431 0.435 0.069
 0.436 0.047 0.022 0.05  0.024 0.08  0.045 0.443 0.065 0.064 0.052 0.45
 0.084 0.448 0.038 0.072 0.459 0.452 0.076 0.433 0.454 0.446 0.447 0.444
 0.071 0.474 0.466 0.499 0.453 0.439 0.528 0.506 0.456 0.44  0.079 0.53
 0.531 0.532 0.485 0.06  0.525 0.494 0.508 0.463 0.522 0.492 0.451 0.471
 0.461 0.545 0.546 0.548 0.549 0.063 0.547 0.437 0.55  0.053 0.054 0.569
 0.068 0.056 0.595 0.061 0.592 0.597 0.587 0.616 0.445 0.251 0.039 0.625
 0.5   0.501 0.638 0.498 0.513 0.641 0.539 0.708 0.644 0.635 0.442 0.68
 0.537 0.672 0.552 0.484 0.766 0.719 0.49  0.683 0.438 0.432 0.744 0.599
 0.519 0.516 0.727 0.458 0.062 0.529 0.769 0.633 0.476 0.64  0.789 0.464
 0.748 0.788 0.489 0.579 0.664 0.774 0.478 0.673 0.472 0.868 0.517 0.856
 0.591 0.479 0.487 0.891 0.455 0.784 0.608 0.505 0.477 0.524 0.468 0.526
 0.611 0.553 0.908 0.805 0.61  0.909 0.475 0.511 0.615 0.9   0.482 0.523
 0.564 0.495 0.623 0.54  0.542 0.823 0.502 0.005 0.898 0.836 0.951 0.602
 0.646 0.874 0.565 0.89  0.665 0.567 0.533 0.873 0.536 0.534 0.753 0.954
 0.563 0.497 0.752 0.491 0.86  0.962 0.883 0.674 0.893 0.981 0.894 0.976
 0.885 0.682 0.967 0.677 0.693 0.556 0.562 0.558 0.47  0.51  0.781 0.928
 0.557 0.578 0.792 0.886 0.889 0.604 0.483 0.607 0.575 0.948 0.923 0.584
 0.628 0.924 0.622 0.624 1.    0.632 0.609]
previous: [0.    0.011 0.004 0.015 0.007 0.04  0.058 0.022 0.018 0.036 0.044 0.025
 0.065 0.033 0.076 0.029 0.051 0.055 0.095 0.135 0.047 0.091 0.073 0.098
 0.062 0.084 0.138 0.105 0.087 0.185 1.    0.08  0.069 0.109 0.211 0.102
 0.116 0.145 0.2   0.127 0.149]
y: [0 1]
job_admin.: [0 1]
job_blue-collar: [0 1]
job_entrepreneur: [0 1]
job_housemaid: [0 1]
job_management: [1 0]
job_retired: [0 1]
job_self-employed: [0 1]
job_services: [0 1]
job_student: [0 1]
job_technician: [0 1]
job_unemployed: [0 1]
job_unknown: [0 1]
education_primary: [0 1]
education_secondary: [0 1]
education_tertiary: [1 0]
education_unknown: [0 1]
marital_divorced: [0 1]
marital_married: [1 0]
marital_single: [0 1]
month_apr: [0 1]
month_aug: [0 1]
month_dec: [0 1]
month_feb: [0 1]
month_jan: [0 1]
month_jul: [0 1]
month_jun: [0 1]
month_mar: [0 1]
month_may: [1 0]
month_nov: [0 1]
month_oct: [0 1]
month_sep: [0 1]
poutcome_failure: [0 1]
poutcome_other: [0 1]
poutcome_success: [0 1]
poutcome_unknown: [1 0]
contact_cellular: [0 1]
contact_telephone: [0 1]
contact_unknown: [1 0]
df.y, df.y.unique()
(0        0
 1        0
 2        0
 3        0
 4        0
         ..
 45206    1
 45207    1
 45208    1
 45209    0
 45210    0
 Name: y, Length: 45211, dtype: int64,
 array([0, 1]))
# split target y from dataframe X
y = df.y
X = df.drop('y',axis='columns')
y
0        0
1        0
2        0
3        0
4        0
        ..
45206    1
45207    1
45208    1
45209    0
45210    0
Name: y, Length: 45211, dtype: int64
X.info()
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 45211 entries, 0 to 45210
Data columns (total 48 columns):
 #   Column               Non-Null Count  Dtype  
---  ------               --------------  -----  
 0   age                  45211 non-null  float64
 1   default              45211 non-null  int64  
 2   balance              45211 non-null  float64
 3   housing              45211 non-null  int64  
 4   loan                 45211 non-null  int64  
 5   day                  45211 non-null  float64
 6   duration             45211 non-null  float64
 7   campaign             45211 non-null  float64
 8   pdays                45211 non-null  float64
 9   previous             45211 non-null  float64
 10  job_admin.           45211 non-null  uint8  
 11  job_blue-collar      45211 non-null  uint8  
 12  job_entrepreneur     45211 non-null  uint8  
 13  job_housemaid        45211 non-null  uint8  
 14  job_management       45211 non-null  uint8  
 15  job_retired          45211 non-null  uint8  
 16  job_self-employed    45211 non-null  uint8  
 17  job_services         45211 non-null  uint8  
 18  job_student          45211 non-null  uint8  
 19  job_technician       45211 non-null  uint8  
 20  job_unemployed       45211 non-null  uint8  
 21  job_unknown          45211 non-null  uint8  
 22  education_primary    45211 non-null  uint8  
 23  education_secondary  45211 non-null  uint8  
 24  education_tertiary   45211 non-null  uint8  
 25  education_unknown    45211 non-null  uint8  
 26  marital_divorced     45211 non-null  uint8  
 27  marital_married      45211 non-null  uint8  
 28  marital_single       45211 non-null  uint8  
 29  month_apr            45211 non-null  uint8  
 30  month_aug            45211 non-null  uint8  
 31  month_dec            45211 non-null  uint8  
 32  month_feb            45211 non-null  uint8  
 33  month_jan            45211 non-null  uint8  
 34  month_jul            45211 non-null  uint8  
 35  month_jun            45211 non-null  uint8  
 36  month_mar            45211 non-null  uint8  
 37  month_may            45211 non-null  uint8  
 38  month_nov            45211 non-null  uint8  
 39  month_oct            45211 non-null  uint8  
 40  month_sep            45211 non-null  uint8  
 41  poutcome_failure     45211 non-null  uint8  
 42  poutcome_other       45211 non-null  uint8  
 43  poutcome_success     45211 non-null  uint8  
 44  poutcome_unknown     45211 non-null  uint8  
 45  contact_cellular     45211 non-null  uint8  
 46  contact_telephone    45211 non-null  uint8  
 47  contact_unknown      45211 non-null  uint8  
dtypes: float64(7), int64(3), uint8(38)
memory usage: 5.1 MB
# prep traning and test data
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=15, stratify=y)
X_train.shape, X_test.shape
((36168, 48), (9043, 48))
y_train.shape, y_test.shape
((36168,), (9043,))
y_train.value_counts(), y.value_counts()
(0    31937
 1     4231
 Name: y, dtype: int64,
 0    39922
 1     5289
 Name: y, dtype: int64)
X_train[:10]
age default balance housing loan day duration campaign pdays previous job_admin. job_blue-collar job_entrepreneur job_housemaid job_management job_retired job_self-employed job_services job_student job_technician job_unemployed job_unknown education_primary education_secondary education_tertiary education_unknown marital_divorced marital_married marital_single month_apr month_aug month_dec month_feb month_jan month_jul month_jun month_mar month_may month_nov month_oct month_sep poutcome_failure poutcome_other poutcome_success poutcome_unknown contact_cellular contact_telephone contact_unknown
33460 0.194805 0 0.134603 0 0 0.633333 0.033144 0.000000 0.000000 0.000000 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 0 0
43272 0.103896 0 0.106668 1 0 0.333333 0.018910 0.000000 0.113532 0.014545 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 1 0 0 1 0 0
5529 0.519481 0 0.072803 1 0 0.733333 0.028060 0.080645 0.000000 0.000000 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 1
11861 0.337662 0 0.075082 0 0 0.633333 0.002847 0.080645 0.000000 0.000000 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0 1
29271 0.142857 0 0.075718 1 1 0.033333 0.020740 0.016129 0.000000 0.000000 0 1 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 1 1 0 0
33020 0.311688 0 0.087992 1 0 0.533333 0.129321 0.032258 0.000000 0.000000 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 0 0
16965 0.532468 0 0.071151 0 1 0.800000 0.031110 0.000000 0.000000 0.000000 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 1 0 0
16975 0.272727 0 0.086349 1 0 0.800000 0.224888 0.000000 0.000000 0.000000 0 0 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 1 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 1 0 0
17911 0.233766 0 0.113486 0 0 0.966667 0.020130 0.016129 0.000000 0.000000 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 0 1 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 1 0
29558 0.194805 0 0.088455 0 0 0.066667 0.051240 0.080645 0.000000 0.000000 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 1 1 0 0
X.head(10)
age default balance housing loan day duration campaign pdays previous job_admin. job_blue-collar job_entrepreneur job_housemaid job_management job_retired job_self-employed job_services job_student job_technician job_unemployed job_unknown education_primary education_secondary education_tertiary education_unknown marital_divorced marital_married marital_single month_apr month_aug month_dec month_feb month_jan month_jul month_jun month_mar month_may month_nov month_oct month_sep poutcome_failure poutcome_other poutcome_success poutcome_unknown contact_cellular contact_telephone contact_unknown
0 0.519481 0 0.092259 1 0 0.133333 0.053070 0.0 0.0 0.0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 1
1 0.337662 0 0.073067 1 0 0.133333 0.030704 0.0 0.0 0.0 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 1
2 0.194805 0 0.072822 1 1 0.133333 0.015453 0.0 0.0 0.0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 1
3 0.376623 0 0.086476 1 0 0.133333 0.018707 0.0 0.0 0.0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 1
4 0.194805 0 0.072812 0 0 0.133333 0.040260 0.0 0.0 0.0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 1
5 0.220779 0 0.074901 1 0 0.133333 0.028264 0.0 0.0 0.0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 1
6 0.129870 0 0.076862 1 1 0.133333 0.044124 0.0 0.0 0.0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 1
7 0.311688 1 0.072822 1 0 0.133333 0.077267 0.0 0.0 0.0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 1 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 1
8 0.519481 0 0.073902 1 0 0.133333 0.010167 0.0 0.0 0.0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 1
9 0.324675 0 0.078187 1 0 0.133333 0.011183 0.0 0.0 0.0 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 1
# count columnsn to determine TF model input shape
len(X_train.columns)
48
# save original df
df_orig = df.copy()
# build tensorflow layers
def ANN(X_train, y_train, X_test, y_test, loss, weights):
    model = keras.Sequential([
        keras.layers.Dense(48, input_dim=48, activation='relu'),
        keras.layers.Dense(24, activation='relu'),
        keras.layers.Dense(1, activation='sigmoid')
    ])

    model.compile(optimizer='adam', loss=loss, metrics=['accuracy'])
    
    if weights == -1:
        model.fit(X_train, y_train, epochs=50)
    else:
        model.fit(X_train, y_train, epochs=50, class_weight = weights)
    
    print(model.evaluate(X_test, y_test))
    
    y_preds = model.predict(X_test)
    y_preds = np.round(y_preds)
    
    print("Classification Report: \n", classification_report(y_test, y_preds))
    
    return y_preds

Train with default imbalanced dataset

y_preds = ANN(X_train, y_train, X_test, y_test, 'binary_crossentropy', -1)
Metal device set to: Apple M1
Epoch 1/50


2022-09-05 11:29:19.946322: I tensorflow/core/common_runtime/pluggable_device/pluggable_device_factory.cc:305] Could not identify NUMA node of platform GPU ID 0, defaulting to 0. Your kernel may not have been built with NUMA support.
2022-09-05 11:29:19.946399: I tensorflow/core/common_runtime/pluggable_device/pluggable_device_factory.cc:271] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 0 MB memory) -> physical PluggableDevice (device: 0, name: METAL, pci bus id: <undefined>)
2022-09-05 11:29:20.045587: W tensorflow/core/platform/profile_utils/cpu_utils.cc:128] Failed to get CPU frequency: 0 Hz


  11/1131 [..............................] - ETA: 6s - loss: 0.5032 - accuracy: 0.8665  

2022-09-05 11:29:20.232674: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:113] Plugin optimizer for device_type GPU is enabled.


1131/1131 [==============================] - 6s 5ms/step - loss: 0.2855 - accuracy: 0.8910
Epoch 2/50
1131/1131 [==============================] - 6s 5ms/step - loss: 0.2261 - accuracy: 0.9023
Epoch 3/50
1131/1131 [==============================] - 5s 5ms/step - loss: 0.2147 - accuracy: 0.9057
Epoch 4/50
1131/1131 [==============================] - 5s 5ms/step - loss: 0.2098 - accuracy: 0.9072
Epoch 5/50
1131/1131 [==============================] - 5s 5ms/step - loss: 0.2056 - accuracy: 0.9082
Epoch 6/50
1131/1131 [==============================] - 5s 5ms/step - loss: 0.2032 - accuracy: 0.9092
Epoch 7/50
1131/1131 [==============================] - 5s 5ms/step - loss: 0.2006 - accuracy: 0.9109
Epoch 8/50
1131/1131 [==============================] - 5s 5ms/step - loss: 0.1980 - accuracy: 0.9119
Epoch 9/50
1131/1131 [==============================] - 5s 5ms/step - loss: 0.1965 - accuracy: 0.9121
Epoch 10/50
1131/1131 [==============================] - 5s 5ms/step - loss: 0.1952 - accuracy: 0.9125
Epoch 11/50
1131/1131 [==============================] - 5s 5ms/step - loss: 0.1929 - accuracy: 0.9140
Epoch 12/50
1131/1131 [==============================] - 5s 5ms/step - loss: 0.1918 - accuracy: 0.9137
Epoch 13/50
1131/1131 [==============================] - 5s 5ms/step - loss: 0.1910 - accuracy: 0.9143
Epoch 14/50
1131/1131 [==============================] - 5s 4ms/step - loss: 0.1894 - accuracy: 0.9150
Epoch 15/50
1131/1131 [==============================] - 5s 4ms/step - loss: 0.1883 - accuracy: 0.9146
Epoch 16/50
1131/1131 [==============================] - 5s 4ms/step - loss: 0.1871 - accuracy: 0.9161
Epoch 17/50
1131/1131 [==============================] - 5s 4ms/step - loss: 0.1861 - accuracy: 0.9165
Epoch 18/50
1131/1131 [==============================] - 5s 5ms/step - loss: 0.1841 - accuracy: 0.9178
Epoch 19/50
1131/1131 [==============================] - 5s 4ms/step - loss: 0.1843 - accuracy: 0.9170
Epoch 20/50
1131/1131 [==============================] - 5s 4ms/step - loss: 0.1834 - accuracy: 0.9183
Epoch 21/50
1131/1131 [==============================] - 5s 4ms/step - loss: 0.1826 - accuracy: 0.9185
Epoch 22/50
1131/1131 [==============================] - 5s 4ms/step - loss: 0.1816 - accuracy: 0.9203
Epoch 23/50
1131/1131 [==============================] - 5s 4ms/step - loss: 0.1815 - accuracy: 0.9190
Epoch 24/50
1131/1131 [==============================] - 5s 4ms/step - loss: 0.1805 - accuracy: 0.9193
Epoch 25/50
1131/1131 [==============================] - 5s 4ms/step - loss: 0.1796 - accuracy: 0.9195
Epoch 26/50
1131/1131 [==============================] - 5s 4ms/step - loss: 0.1789 - accuracy: 0.9207
Epoch 27/50
1131/1131 [==============================] - 5s 4ms/step - loss: 0.1776 - accuracy: 0.9223
Epoch 28/50
1131/1131 [==============================] - 5s 4ms/step - loss: 0.1776 - accuracy: 0.9217
Epoch 29/50
1131/1131 [==============================] - 5s 4ms/step - loss: 0.1771 - accuracy: 0.9214
Epoch 30/50
1131/1131 [==============================] - 5s 4ms/step - loss: 0.1755 - accuracy: 0.9224
Epoch 31/50
1131/1131 [==============================] - 5s 4ms/step - loss: 0.1744 - accuracy: 0.9232
Epoch 32/50
1131/1131 [==============================] - 5s 4ms/step - loss: 0.1746 - accuracy: 0.9226
Epoch 33/50
1131/1131 [==============================] - 5s 4ms/step - loss: 0.1735 - accuracy: 0.9238
Epoch 34/50
1131/1131 [==============================] - 5s 4ms/step - loss: 0.1730 - accuracy: 0.9244
Epoch 35/50
1131/1131 [==============================] - 5s 4ms/step - loss: 0.1715 - accuracy: 0.9249
Epoch 36/50
1131/1131 [==============================] - 5s 4ms/step - loss: 0.1708 - accuracy: 0.9256
Epoch 37/50
1131/1131 [==============================] - 5s 4ms/step - loss: 0.1709 - accuracy: 0.9244
Epoch 38/50
1131/1131 [==============================] - 5s 4ms/step - loss: 0.1698 - accuracy: 0.9265
Epoch 39/50
1131/1131 [==============================] - 5s 4ms/step - loss: 0.1695 - accuracy: 0.9260
Epoch 40/50
1131/1131 [==============================] - 5s 4ms/step - loss: 0.1686 - accuracy: 0.9263
Epoch 41/50
1131/1131 [==============================] - 5s 4ms/step - loss: 0.1686 - accuracy: 0.9274
Epoch 42/50
1131/1131 [==============================] - 5s 4ms/step - loss: 0.1680 - accuracy: 0.9267
Epoch 43/50
1131/1131 [==============================] - 5s 4ms/step - loss: 0.1672 - accuracy: 0.9267
Epoch 44/50
1131/1131 [==============================] - 5s 4ms/step - loss: 0.1669 - accuracy: 0.9265
Epoch 45/50
1131/1131 [==============================] - 5s 4ms/step - loss: 0.1655 - accuracy: 0.9277
Epoch 46/50
1131/1131 [==============================] - 5s 4ms/step - loss: 0.1653 - accuracy: 0.9279
Epoch 47/50
1131/1131 [==============================] - 5s 4ms/step - loss: 0.1648 - accuracy: 0.9272
Epoch 48/50
1131/1131 [==============================] - 5s 4ms/step - loss: 0.1644 - accuracy: 0.9294
Epoch 49/50
1131/1131 [==============================] - 5s 4ms/step - loss: 0.1637 - accuracy: 0.9287
Epoch 50/50
1131/1131 [==============================] - 5s 4ms/step - loss: 0.1627 - accuracy: 0.9288
 52/283 [====>.........................] - ETA: 0s - loss: 0.2444 - accuracy: 0.8972

2022-09-05 11:33:35.899244: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:113] Plugin optimizer for device_type GPU is enabled.


283/283 [==============================] - 1s 3ms/step - loss: 0.2238 - accuracy: 0.9016
[0.2238490879535675, 0.9015812873840332]
120/283 [===========>..................] - ETA: 0s

2022-09-05 11:33:36.840196: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:113] Plugin optimizer for device_type GPU is enabled.


283/283 [==============================] - 0s 1ms/step
Classification Report: 
               precision    recall  f1-score   support

           0       0.93      0.96      0.94      7985
           1       0.60      0.49      0.54      1058

    accuracy                           0.90      9043
   macro avg       0.77      0.72      0.74      9043
weighted avg       0.89      0.90      0.90      9043

Train by Undersampling Majority Class

df = df_orig

count_class_0, count_class_1 = df.y.value_counts()
# Divide by class
df_class_0 = df[df['y'] == 0]
df_class_1 = df[df['y'] == 1]
# Undersample 0 class in y and combine DataFrames of both class 0 and 1
df_class_0_under = df_class_0.sample(count_class_1)
df_test_under = pd.concat([df_class_0_under, df_class_1], axis=0)

print('Random undersampling:')
print(df_test_under.y.value_counts())
Random undersampling:
0    5289
1    5289
Name: y, dtype: int64
X = df_test_under.drop('y',axis='columns')
y = df_test_under['y']

from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=15, stratify=y)
# Number of classes in training Data
y_train.value_counts()
0    4231
1    4231
Name: y, dtype: int64
y_preds = ANN(X_train, y_train, X_test, y_test, 'binary_crossentropy', -1)
Epoch 1/50
 23/265 [=>............................] - ETA: 1s - loss: 0.6910 - accuracy: 0.5245

2022-09-05 11:33:37.467849: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:113] Plugin optimizer for device_type GPU is enabled.


265/265 [==============================] - 1s 5ms/step - loss: 0.6106 - accuracy: 0.6615
Epoch 2/50
265/265 [==============================] - 1s 4ms/step - loss: 0.5071 - accuracy: 0.7681
Epoch 3/50
265/265 [==============================] - 1s 4ms/step - loss: 0.4339 - accuracy: 0.8175
Epoch 4/50
265/265 [==============================] - 1s 4ms/step - loss: 0.3899 - accuracy: 0.8375
Epoch 5/50
265/265 [==============================] - 1s 5ms/step - loss: 0.3715 - accuracy: 0.8416
Epoch 6/50
265/265 [==============================] - 1s 4ms/step - loss: 0.3595 - accuracy: 0.8504
Epoch 7/50
265/265 [==============================] - 1s 4ms/step - loss: 0.3515 - accuracy: 0.8513
Epoch 8/50
265/265 [==============================] - 1s 5ms/step - loss: 0.3438 - accuracy: 0.8563
Epoch 9/50
265/265 [==============================] - 1s 5ms/step - loss: 0.3384 - accuracy: 0.8559
Epoch 10/50
265/265 [==============================] - 1s 5ms/step - loss: 0.3349 - accuracy: 0.8589
Epoch 11/50
265/265 [==============================] - 1s 4ms/step - loss: 0.3296 - accuracy: 0.8659
Epoch 12/50
265/265 [==============================] - 1s 4ms/step - loss: 0.3245 - accuracy: 0.8659
Epoch 13/50
265/265 [==============================] - 1s 4ms/step - loss: 0.3237 - accuracy: 0.8679
Epoch 14/50
265/265 [==============================] - 1s 5ms/step - loss: 0.3166 - accuracy: 0.8702
Epoch 15/50
265/265 [==============================] - 1s 4ms/step - loss: 0.3136 - accuracy: 0.8727
Epoch 16/50
265/265 [==============================] - 1s 4ms/step - loss: 0.3098 - accuracy: 0.8733
Epoch 17/50
265/265 [==============================] - 1s 4ms/step - loss: 0.3067 - accuracy: 0.8751
Epoch 18/50
265/265 [==============================] - 1s 4ms/step - loss: 0.3073 - accuracy: 0.8730
Epoch 19/50
265/265 [==============================] - 1s 5ms/step - loss: 0.3031 - accuracy: 0.8744
Epoch 20/50
265/265 [==============================] - 1s 5ms/step - loss: 0.2999 - accuracy: 0.8775
Epoch 21/50
265/265 [==============================] - 1s 4ms/step - loss: 0.2981 - accuracy: 0.8779
Epoch 22/50
265/265 [==============================] - 1s 4ms/step - loss: 0.2940 - accuracy: 0.8817
Epoch 23/50
265/265 [==============================] - 1s 4ms/step - loss: 0.2950 - accuracy: 0.8802
Epoch 24/50
265/265 [==============================] - 1s 4ms/step - loss: 0.2896 - accuracy: 0.8808
Epoch 25/50
265/265 [==============================] - 1s 4ms/step - loss: 0.2900 - accuracy: 0.8816
Epoch 26/50
265/265 [==============================] - 1s 5ms/step - loss: 0.2881 - accuracy: 0.8827
Epoch 27/50
265/265 [==============================] - 1s 4ms/step - loss: 0.2860 - accuracy: 0.8838
Epoch 28/50
265/265 [==============================] - 1s 4ms/step - loss: 0.2838 - accuracy: 0.8851
Epoch 29/50
265/265 [==============================] - 1s 4ms/step - loss: 0.2858 - accuracy: 0.8818
Epoch 30/50
265/265 [==============================] - 1s 5ms/step - loss: 0.2810 - accuracy: 0.8827
Epoch 31/50
265/265 [==============================] - 1s 4ms/step - loss: 0.2794 - accuracy: 0.8860
Epoch 32/50
265/265 [==============================] - 1s 5ms/step - loss: 0.2777 - accuracy: 0.8874
Epoch 33/50
265/265 [==============================] - 1s 5ms/step - loss: 0.2750 - accuracy: 0.8875
Epoch 34/50
265/265 [==============================] - 1s 4ms/step - loss: 0.2742 - accuracy: 0.8882
Epoch 35/50
265/265 [==============================] - 1s 4ms/step - loss: 0.2730 - accuracy: 0.8893
Epoch 36/50
265/265 [==============================] - 1s 4ms/step - loss: 0.2719 - accuracy: 0.8877
Epoch 37/50
265/265 [==============================] - 1s 4ms/step - loss: 0.2704 - accuracy: 0.8897
Epoch 38/50
265/265 [==============================] - 1s 4ms/step - loss: 0.2661 - accuracy: 0.8928
Epoch 39/50
265/265 [==============================] - 1s 4ms/step - loss: 0.2687 - accuracy: 0.8892
Epoch 40/50
265/265 [==============================] - 1s 4ms/step - loss: 0.2658 - accuracy: 0.8922
Epoch 41/50
265/265 [==============================] - 1s 4ms/step - loss: 0.2650 - accuracy: 0.8914
Epoch 42/50
265/265 [==============================] - 1s 5ms/step - loss: 0.2640 - accuracy: 0.8932
Epoch 43/50
265/265 [==============================] - 1s 4ms/step - loss: 0.2613 - accuracy: 0.8940
Epoch 44/50
265/265 [==============================] - 1s 4ms/step - loss: 0.2610 - accuracy: 0.8922
Epoch 45/50
265/265 [==============================] - 1s 4ms/step - loss: 0.2603 - accuracy: 0.8927
Epoch 46/50
265/265 [==============================] - 1s 4ms/step - loss: 0.2590 - accuracy: 0.8932
Epoch 47/50
265/265 [==============================] - 1s 4ms/step - loss: 0.2574 - accuracy: 0.8938
Epoch 48/50
265/265 [==============================] - 1s 5ms/step - loss: 0.2573 - accuracy: 0.8936
Epoch 49/50
265/265 [==============================] - 1s 5ms/step - loss: 0.2522 - accuracy: 0.8972
Epoch 50/50
265/265 [==============================] - 1s 5ms/step - loss: 0.2544 - accuracy: 0.8935
51/67 [=====================>........] - ETA: 0s - loss: 0.3816 - accuracy: 0.8554

2022-09-05 11:34:37.386744: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:113] Plugin optimizer for device_type GPU is enabled.


67/67 [==============================] - 0s 3ms/step - loss: 0.3852 - accuracy: 0.8521
[0.38524705171585083, 0.8520793914794922]
67/67 [==============================] - 0s 1ms/step
Classification Report: 
               precision    recall  f1-score   support

           0       0.88      0.82      0.85      1058
           1       0.83      0.89      0.86      1058

    accuracy                           0.85      2116
   macro avg       0.85      0.85      0.85      2116
weighted avg       0.85      0.85      0.85      2116



2022-09-05 11:34:37.686646: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:113] Plugin optimizer for device_type GPU is enabled.
# The F1 score results are much better than training with default dataset. 

Train by Oversampling Minority Class

df = df_orig

# Oversample 1-class and concat the DataFrames of both classes
df_class_1_over = df_class_1.sample(count_class_0, replace=True)
df_test_over = pd.concat([df_class_0, df_class_1_over], axis=0)

print('Random over-sampling:')
print(df_test_over.y.value_counts())
Random over-sampling:
0    39922
1    39922
Name: y, dtype: int64
# split X and y

X = df_test_over.drop('y',axis='columns')
y = df_test_over['y']

from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=15, stratify=y)
# Number of classes in training Data
y_train.value_counts()
1    31938
0    31937
Name: y, dtype: int64
loss = keras.losses.BinaryCrossentropy()
weights = -1
y_preds = ANN(X_train, y_train, X_test, y_test, 'binary_crossentropy', -1)
Epoch 1/50
  22/1997 [..............................] - ETA: 9s - loss: 0.7002 - accuracy: 0.5170 

2022-09-05 11:34:38.043546: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:113] Plugin optimizer for device_type GPU is enabled.


1997/1997 [==============================] - 9s 4ms/step - loss: 0.4215 - accuracy: 0.8104
Epoch 2/50
1997/1997 [==============================] - 9s 4ms/step - loss: 0.3336 - accuracy: 0.8615
Epoch 3/50
1997/1997 [==============================] - 9s 4ms/step - loss: 0.3175 - accuracy: 0.8694
Epoch 4/50
1997/1997 [==============================] - 9s 4ms/step - loss: 0.3073 - accuracy: 0.8747
Epoch 5/50
1997/1997 [==============================] - 9s 4ms/step - loss: 0.3000 - accuracy: 0.8773
Epoch 6/50
1997/1997 [==============================] - 9s 4ms/step - loss: 0.2938 - accuracy: 0.8800
Epoch 7/50
1997/1997 [==============================] - 9s 4ms/step - loss: 0.2900 - accuracy: 0.8830
Epoch 8/50
1997/1997 [==============================] - 9s 4ms/step - loss: 0.2866 - accuracy: 0.8831
Epoch 9/50
1997/1997 [==============================] - 9s 4ms/step - loss: 0.2827 - accuracy: 0.8860
Epoch 10/50
1997/1997 [==============================] - 9s 4ms/step - loss: 0.2792 - accuracy: 0.8873
Epoch 11/50
1997/1997 [==============================] - 9s 4ms/step - loss: 0.2751 - accuracy: 0.8888
Epoch 12/50
1997/1997 [==============================] - 9s 4ms/step - loss: 0.2732 - accuracy: 0.8891
Epoch 13/50
1997/1997 [==============================] - 9s 4ms/step - loss: 0.2709 - accuracy: 0.8895
Epoch 14/50
1997/1997 [==============================] - 9s 4ms/step - loss: 0.2675 - accuracy: 0.8921
Epoch 15/50
1997/1997 [==============================] - 9s 4ms/step - loss: 0.2660 - accuracy: 0.8920
Epoch 16/50
1997/1997 [==============================] - 9s 4ms/step - loss: 0.2636 - accuracy: 0.8934
Epoch 17/50
1997/1997 [==============================] - 9s 5ms/step - loss: 0.2610 - accuracy: 0.8950
Epoch 18/50
1997/1997 [==============================] - 9s 5ms/step - loss: 0.2588 - accuracy: 0.8962
Epoch 19/50
1997/1997 [==============================] - 9s 5ms/step - loss: 0.2571 - accuracy: 0.8966
Epoch 20/50
1997/1997 [==============================] - 9s 5ms/step - loss: 0.2553 - accuracy: 0.8977
Epoch 21/50
1997/1997 [==============================] - 9s 5ms/step - loss: 0.2544 - accuracy: 0.8980
Epoch 22/50
1997/1997 [==============================] - 9s 4ms/step - loss: 0.2516 - accuracy: 0.8990
Epoch 23/50
1997/1997 [==============================] - 9s 5ms/step - loss: 0.2506 - accuracy: 0.8997
Epoch 24/50
1997/1997 [==============================] - 9s 5ms/step - loss: 0.2497 - accuracy: 0.9006
Epoch 25/50
1997/1997 [==============================] - 9s 5ms/step - loss: 0.2475 - accuracy: 0.9018
Epoch 26/50
1997/1997 [==============================] - 9s 5ms/step - loss: 0.2460 - accuracy: 0.9018
Epoch 27/50
1997/1997 [==============================] - 9s 5ms/step - loss: 0.2447 - accuracy: 0.9028
Epoch 28/50
1997/1997 [==============================] - 9s 5ms/step - loss: 0.2427 - accuracy: 0.9040
Epoch 29/50
1997/1997 [==============================] - 9s 5ms/step - loss: 0.2412 - accuracy: 0.9046
Epoch 30/50
1997/1997 [==============================] - 9s 5ms/step - loss: 0.2399 - accuracy: 0.9051
Epoch 31/50
1997/1997 [==============================] - 9s 5ms/step - loss: 0.2390 - accuracy: 0.9057
Epoch 32/50
1997/1997 [==============================] - 9s 5ms/step - loss: 0.2381 - accuracy: 0.9060
Epoch 33/50
1997/1997 [==============================] - 9s 5ms/step - loss: 0.2358 - accuracy: 0.9082
Epoch 34/50
1997/1997 [==============================] - 9s 5ms/step - loss: 0.2354 - accuracy: 0.9064
Epoch 35/50
1997/1997 [==============================] - 9s 5ms/step - loss: 0.2332 - accuracy: 0.9083
Epoch 36/50
1997/1997 [==============================] - 9s 5ms/step - loss: 0.2329 - accuracy: 0.9085
Epoch 37/50
1997/1997 [==============================] - 9s 5ms/step - loss: 0.2313 - accuracy: 0.9099
Epoch 38/50
1997/1997 [==============================] - 9s 4ms/step - loss: 0.2296 - accuracy: 0.9109
Epoch 39/50
1997/1997 [==============================] - 9s 4ms/step - loss: 0.2288 - accuracy: 0.9111
Epoch 40/50
1997/1997 [==============================] - 9s 4ms/step - loss: 0.2284 - accuracy: 0.9111
Epoch 41/50
1997/1997 [==============================] - 9s 4ms/step - loss: 0.2275 - accuracy: 0.9119
Epoch 42/50
1997/1997 [==============================] - 9s 5ms/step - loss: 0.2262 - accuracy: 0.9127
Epoch 43/50
1997/1997 [==============================] - 10s 5ms/step - loss: 0.2253 - accuracy: 0.9134
Epoch 44/50
1997/1997 [==============================] - 710s 356ms/step - loss: 0.2243 - accuracy: 0.9139
Epoch 45/50
1997/1997 [==============================] - 9s 5ms/step - loss: 0.2227 - accuracy: 0.9139
Epoch 46/50
1997/1997 [==============================] - 9s 4ms/step - loss: 0.2217 - accuracy: 0.9150
Epoch 47/50
1997/1997 [==============================] - 9s 4ms/step - loss: 0.2217 - accuracy: 0.9144
Epoch 48/50
1997/1997 [==============================] - 9s 4ms/step - loss: 0.2203 - accuracy: 0.9155
Epoch 49/50
1997/1997 [==============================] - 9s 4ms/step - loss: 0.2196 - accuracy: 0.9163
Epoch 50/50
1997/1997 [==============================] - 9s 4ms/step - loss: 0.2189 - accuracy: 0.9152
 50/500 [==>...........................] - ETA: 1s - loss: 0.2629 - accuracy: 0.9094

2022-09-05 11:53:50.099041: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:113] Plugin optimizer for device_type GPU is enabled.


500/500 [==============================] - 2s 3ms/step - loss: 0.2614 - accuracy: 0.9022
[0.2613811194896698, 0.9022481441497803]
122/500 [======>.......................] - ETA: 0s

2022-09-05 11:53:51.702137: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:113] Plugin optimizer for device_type GPU is enabled.


500/500 [==============================] - 1s 1ms/step
Classification Report: 
               precision    recall  f1-score   support

           0       0.94      0.86      0.90      7985
           1       0.87      0.94      0.91      7984

    accuracy                           0.90     15969
   macro avg       0.90      0.90      0.90     15969
weighted avg       0.90      0.90      0.90     15969
# The F1 score results are better than training with either default dataset and with undersmapling. 

Train by Undersampling via Ensemble

df = df_orig
df.y.value_counts()
0    39922
1     5289
Name: y, dtype: int64
# split into three undersampled sets

# 0    39922
# 1     5289

# model1 --> class1(5289) + class0(0, 5289)
# model2 --> class1(5289) + class0(5290, 10579)
# model3 --> class1(5289) + class0(10579, 15868)
# rebuild X y split
X = df.drop('y',axis='columns')
y = df['y']
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=15, stratify=y)
y_train.value_counts()
0    31937
1     4231
Name: y, dtype: int64
df = X_train.copy()
df['y'] = y_train
df.head()
age default balance housing loan day duration campaign pdays previous job_admin. job_blue-collar job_entrepreneur job_housemaid job_management job_retired job_self-employed job_services job_student job_technician job_unemployed job_unknown education_primary education_secondary education_tertiary education_unknown marital_divorced marital_married marital_single month_apr month_aug month_dec month_feb month_jan month_jul month_jun month_mar month_may month_nov month_oct month_sep poutcome_failure poutcome_other poutcome_success poutcome_unknown contact_cellular contact_telephone contact_unknown y
33460 0.194805 0 0.134603 0 0 0.633333 0.033144 0.000000 0.000000 0.000000 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 0 0 1
43272 0.103896 0 0.106668 1 0 0.333333 0.018910 0.000000 0.113532 0.014545 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 1 0 0 1 0 0 0
5529 0.519481 0 0.072803 1 0 0.733333 0.028060 0.080645 0.000000 0.000000 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 1 0
11861 0.337662 0 0.075082 0 0 0.633333 0.002847 0.080645 0.000000 0.000000 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0 1 0
29271 0.142857 0 0.075718 1 1 0.033333 0.020740 0.016129 0.000000 0.000000 0 1 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 1 1 0 0 0
df_class0 = df[df.y==0]
df_class1 = df[df.y==1]
def get_train_batch(df_majority, df_minority, start, end):
    df_train = pd.concat([df_majority[start:end], df_minority], axis=0)

    X_train = df_train.drop('y', axis='columns')
    y_train = df_train.y
    return X_train, y_train    
# train the first model

X_train, y_train = get_train_batch(df_class0, df_class1, 0, 5289)
y_pred1 = ANN(X_train, y_train, X_test, y_test, 'binary_crossentropy', -1)
Epoch 1/50
 23/298 [=>............................] - ETA: 1s - loss: 0.6727 - accuracy: 0.5870

2022-09-05 11:53:52.655998: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:113] Plugin optimizer for device_type GPU is enabled.


298/298 [==============================] - 2s 5ms/step - loss: 0.5743 - accuracy: 0.7086
Epoch 2/50
298/298 [==============================] - 1s 4ms/step - loss: 0.4729 - accuracy: 0.7836
Epoch 3/50
298/298 [==============================] - 1s 5ms/step - loss: 0.4080 - accuracy: 0.8202
Epoch 4/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3807 - accuracy: 0.8351
Epoch 5/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3652 - accuracy: 0.8408
Epoch 6/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3573 - accuracy: 0.8487
Epoch 7/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3497 - accuracy: 0.8511
Epoch 8/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3416 - accuracy: 0.8563
Epoch 9/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3351 - accuracy: 0.8565
Epoch 10/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3316 - accuracy: 0.8612
Epoch 11/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3242 - accuracy: 0.8636
Epoch 12/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3215 - accuracy: 0.8629
Epoch 13/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3163 - accuracy: 0.8689
Epoch 14/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3127 - accuracy: 0.8684
Epoch 15/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3099 - accuracy: 0.8701
Epoch 16/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3057 - accuracy: 0.8706
Epoch 17/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3024 - accuracy: 0.8738
Epoch 18/50
298/298 [==============================] - 1s 4ms/step - loss: 0.2986 - accuracy: 0.8716
Epoch 19/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2969 - accuracy: 0.8738
Epoch 20/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2969 - accuracy: 0.8757
Epoch 21/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2919 - accuracy: 0.8780
Epoch 22/50
298/298 [==============================] - 1s 4ms/step - loss: 0.2888 - accuracy: 0.8786
Epoch 23/50
298/298 [==============================] - 1s 4ms/step - loss: 0.2911 - accuracy: 0.8768
Epoch 24/50
298/298 [==============================] - 1s 4ms/step - loss: 0.2861 - accuracy: 0.8798
Epoch 25/50
298/298 [==============================] - 1s 4ms/step - loss: 0.2841 - accuracy: 0.8799
Epoch 26/50
298/298 [==============================] - 1s 4ms/step - loss: 0.2832 - accuracy: 0.8807
Epoch 27/50
298/298 [==============================] - 1s 4ms/step - loss: 0.2823 - accuracy: 0.8851
Epoch 28/50
298/298 [==============================] - 1s 4ms/step - loss: 0.2774 - accuracy: 0.8852
Epoch 29/50
298/298 [==============================] - 1s 4ms/step - loss: 0.2763 - accuracy: 0.8840
Epoch 30/50
298/298 [==============================] - 1s 4ms/step - loss: 0.2743 - accuracy: 0.8856
Epoch 31/50
298/298 [==============================] - 1s 4ms/step - loss: 0.2752 - accuracy: 0.8837
Epoch 32/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2737 - accuracy: 0.8845
Epoch 33/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2721 - accuracy: 0.8852
Epoch 34/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2708 - accuracy: 0.8859
Epoch 35/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2686 - accuracy: 0.8899
Epoch 36/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2665 - accuracy: 0.8888
Epoch 37/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2663 - accuracy: 0.8873
Epoch 38/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2657 - accuracy: 0.8880
Epoch 39/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2628 - accuracy: 0.8929
Epoch 40/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2615 - accuracy: 0.8903
Epoch 41/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2624 - accuracy: 0.8894
Epoch 42/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2600 - accuracy: 0.8932
Epoch 43/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2579 - accuracy: 0.8933
Epoch 44/50
298/298 [==============================] - 1s 4ms/step - loss: 0.2586 - accuracy: 0.8912
Epoch 45/50
298/298 [==============================] - 1s 4ms/step - loss: 0.2562 - accuracy: 0.8917
Epoch 46/50
298/298 [==============================] - 1s 4ms/step - loss: 0.2542 - accuracy: 0.8954
Epoch 47/50
298/298 [==============================] - 1s 4ms/step - loss: 0.2547 - accuracy: 0.8949
Epoch 48/50
298/298 [==============================] - 1s 4ms/step - loss: 0.2539 - accuracy: 0.8959
Epoch 49/50
298/298 [==============================] - 1s 4ms/step - loss: 0.2559 - accuracy: 0.8932
Epoch 50/50
298/298 [==============================] - 1s 4ms/step - loss: 0.2505 - accuracy: 0.8968
 48/283 [====>.........................] - ETA: 0s - loss: 0.4895 - accuracy: 0.7923

2022-09-05 11:55:00.887363: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:113] Plugin optimizer for device_type GPU is enabled.


283/283 [==============================] - 1s 3ms/step - loss: 0.4620 - accuracy: 0.7975
[0.4620147943496704, 0.7975229024887085]
121/283 [===========>..................] - ETA: 0s

2022-09-05 11:55:02.005051: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:113] Plugin optimizer for device_type GPU is enabled.


283/283 [==============================] - 0s 1ms/step
Classification Report: 
               precision    recall  f1-score   support

           0       0.99      0.78      0.87      7985
           1       0.36      0.91      0.51      1058

    accuracy                           0.80      9043
   macro avg       0.67      0.85      0.69      9043
weighted avg       0.91      0.80      0.83      9043
# train the second model

X_train, y_train = get_train_batch(df_class0, df_class1, 5290, 10579)
y_pred2 = ANN(X_train, y_train, X_test, y_test, 'binary_crossentropy', -1)
Epoch 1/50
 22/298 [=>............................] - ETA: 1s - loss: 0.6829 - accuracy: 0.5412

2022-09-05 11:55:02.600109: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:113] Plugin optimizer for device_type GPU is enabled.


298/298 [==============================] - 2s 5ms/step - loss: 0.5872 - accuracy: 0.6886
Epoch 2/50
298/298 [==============================] - 1s 5ms/step - loss: 0.4697 - accuracy: 0.7853
Epoch 3/50
298/298 [==============================] - 1s 5ms/step - loss: 0.4002 - accuracy: 0.8273
Epoch 4/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3734 - accuracy: 0.8388
Epoch 5/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3572 - accuracy: 0.8455
Epoch 6/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3499 - accuracy: 0.8494
Epoch 7/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3425 - accuracy: 0.8539
Epoch 8/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3346 - accuracy: 0.8546
Epoch 9/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3307 - accuracy: 0.8584
Epoch 10/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3260 - accuracy: 0.8589
Epoch 11/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3179 - accuracy: 0.8640
Epoch 12/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3131 - accuracy: 0.8657
Epoch 13/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3107 - accuracy: 0.8664
Epoch 14/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3086 - accuracy: 0.8680
Epoch 15/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3043 - accuracy: 0.8723
Epoch 16/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2987 - accuracy: 0.8723
Epoch 17/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2979 - accuracy: 0.8750
Epoch 18/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2930 - accuracy: 0.8758
Epoch 19/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2921 - accuracy: 0.8769
Epoch 20/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2901 - accuracy: 0.8790
Epoch 21/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2881 - accuracy: 0.8800
Epoch 22/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2835 - accuracy: 0.8779
Epoch 23/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2838 - accuracy: 0.8817
Epoch 24/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2791 - accuracy: 0.8824
Epoch 25/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2804 - accuracy: 0.8815
Epoch 26/50
298/298 [==============================] - 1s 4ms/step - loss: 0.2782 - accuracy: 0.8816
Epoch 27/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2767 - accuracy: 0.8862
Epoch 28/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2738 - accuracy: 0.8859
Epoch 29/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2723 - accuracy: 0.8848
Epoch 30/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2699 - accuracy: 0.8888
Epoch 31/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2697 - accuracy: 0.8882
Epoch 32/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2668 - accuracy: 0.8895
Epoch 33/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2654 - accuracy: 0.8903
Epoch 34/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2648 - accuracy: 0.8890
Epoch 35/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2654 - accuracy: 0.8860
Epoch 36/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2622 - accuracy: 0.8920
Epoch 37/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2607 - accuracy: 0.8905
Epoch 38/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2606 - accuracy: 0.8914
Epoch 39/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2574 - accuracy: 0.8925
Epoch 40/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2572 - accuracy: 0.8926
Epoch 41/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2541 - accuracy: 0.8959
Epoch 42/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2537 - accuracy: 0.8947
Epoch 43/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2511 - accuracy: 0.8960
Epoch 44/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2514 - accuracy: 0.8959
Epoch 45/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2482 - accuracy: 0.8974
Epoch 46/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2496 - accuracy: 0.8975
Epoch 47/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2472 - accuracy: 0.8965
Epoch 48/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2463 - accuracy: 0.8981
Epoch 49/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2479 - accuracy: 0.8980
Epoch 50/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2459 - accuracy: 0.8983
 50/283 [====>.........................] - ETA: 0s - loss: 0.4942 - accuracy: 0.7919

2022-09-05 11:56:11.767969: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:113] Plugin optimizer for device_type GPU is enabled.


283/283 [==============================] - 1s 3ms/step - loss: 0.4857 - accuracy: 0.7945
[0.48568180203437805, 0.7945371866226196]
123/283 [============>.................] - ETA: 0s

2022-09-05 11:56:12.711325: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:113] Plugin optimizer for device_type GPU is enabled.


283/283 [==============================] - 0s 1ms/step
Classification Report: 
               precision    recall  f1-score   support

           0       0.98      0.78      0.87      7985
           1       0.35      0.91      0.51      1058

    accuracy                           0.79      9043
   macro avg       0.67      0.84      0.69      9043
weighted avg       0.91      0.79      0.83      9043
# train the third model

X_train, y_train = get_train_batch(df_class0, df_class1, 10579, 15868)

y_pred3 = ANN(X_train, y_train, X_test, y_test, 'binary_crossentropy', -1)
Epoch 1/50
 22/298 [=>............................] - ETA: 1s - loss: 0.6776 - accuracy: 0.5653

2022-09-05 11:56:13.304061: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:113] Plugin optimizer for device_type GPU is enabled.


298/298 [==============================] - 2s 5ms/step - loss: 0.5789 - accuracy: 0.7042
Epoch 2/50
298/298 [==============================] - 1s 5ms/step - loss: 0.4759 - accuracy: 0.7818
Epoch 3/50
298/298 [==============================] - 1s 5ms/step - loss: 0.4085 - accuracy: 0.8224
Epoch 4/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3777 - accuracy: 0.8368
Epoch 5/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3618 - accuracy: 0.8454
Epoch 6/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3531 - accuracy: 0.8494
Epoch 7/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3451 - accuracy: 0.8515
Epoch 8/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3390 - accuracy: 0.8528
Epoch 9/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3331 - accuracy: 0.8618
Epoch 10/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3298 - accuracy: 0.8599
Epoch 11/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3234 - accuracy: 0.8621
Epoch 12/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3178 - accuracy: 0.8649
Epoch 13/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3157 - accuracy: 0.8667
Epoch 14/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3110 - accuracy: 0.8711
Epoch 15/50
298/298 [==============================] - 1s 4ms/step - loss: 0.3097 - accuracy: 0.8654
Epoch 16/50
298/298 [==============================] - 1s 4ms/step - loss: 0.3067 - accuracy: 0.8679
Epoch 17/50
298/298 [==============================] - 1s 4ms/step - loss: 0.3035 - accuracy: 0.8709
Epoch 18/50
298/298 [==============================] - 1s 5ms/step - loss: 0.3007 - accuracy: 0.8729
Epoch 19/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2997 - accuracy: 0.8718
Epoch 20/50
298/298 [==============================] - 1s 4ms/step - loss: 0.2996 - accuracy: 0.8726
Epoch 21/50
298/298 [==============================] - 1s 4ms/step - loss: 0.2950 - accuracy: 0.8731
Epoch 22/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2915 - accuracy: 0.8745
Epoch 23/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2906 - accuracy: 0.8758
Epoch 24/50
298/298 [==============================] - 1s 4ms/step - loss: 0.2886 - accuracy: 0.8795
Epoch 25/50
298/298 [==============================] - 1s 4ms/step - loss: 0.2869 - accuracy: 0.8782
Epoch 26/50
298/298 [==============================] - 1s 4ms/step - loss: 0.2838 - accuracy: 0.8813
Epoch 27/50
298/298 [==============================] - 1s 4ms/step - loss: 0.2833 - accuracy: 0.8812
Epoch 28/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2824 - accuracy: 0.8812
Epoch 29/50
298/298 [==============================] - 1s 4ms/step - loss: 0.2789 - accuracy: 0.8832
Epoch 30/50
298/298 [==============================] - 1s 4ms/step - loss: 0.2767 - accuracy: 0.8830
Epoch 31/50
298/298 [==============================] - 1s 4ms/step - loss: 0.2752 - accuracy: 0.8822
Epoch 32/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2737 - accuracy: 0.8848
Epoch 33/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2730 - accuracy: 0.8848
Epoch 34/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2705 - accuracy: 0.8846
Epoch 35/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2688 - accuracy: 0.8859
Epoch 36/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2656 - accuracy: 0.8895
Epoch 37/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2657 - accuracy: 0.8893
Epoch 38/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2656 - accuracy: 0.8883
Epoch 39/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2626 - accuracy: 0.8879
Epoch 40/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2606 - accuracy: 0.8893
Epoch 41/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2622 - accuracy: 0.8891
Epoch 42/50
298/298 [==============================] - 2s 5ms/step - loss: 0.2575 - accuracy: 0.8945
Epoch 43/50
298/298 [==============================] - 2s 5ms/step - loss: 0.2547 - accuracy: 0.8928
Epoch 44/50
298/298 [==============================] - 2s 5ms/step - loss: 0.2538 - accuracy: 0.8954
Epoch 45/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2537 - accuracy: 0.8939
Epoch 46/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2505 - accuracy: 0.8956
Epoch 47/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2510 - accuracy: 0.8950
Epoch 48/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2511 - accuracy: 0.8964
Epoch 49/50
298/298 [==============================] - 2s 5ms/step - loss: 0.2519 - accuracy: 0.8929
Epoch 50/50
298/298 [==============================] - 1s 5ms/step - loss: 0.2491 - accuracy: 0.8935
 47/283 [===>..........................] - ETA: 0s - loss: 0.5068 - accuracy: 0.7972

2022-09-05 11:57:23.078053: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:113] Plugin optimizer for device_type GPU is enabled.


283/283 [==============================] - 1s 3ms/step - loss: 0.4903 - accuracy: 0.7897
[0.49025654792785645, 0.7896715402603149]
111/283 [==========>...................] - ETA: 0s

2022-09-05 11:57:24.090148: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:113] Plugin optimizer for device_type GPU is enabled.


283/283 [==============================] - 0s 1ms/step
Classification Report: 
               precision    recall  f1-score   support

           0       0.99      0.77      0.87      7985
           1       0.35      0.92      0.50      1058

    accuracy                           0.79      9043
   macro avg       0.67      0.84      0.69      9043
weighted avg       0.91      0.79      0.82      9043
# combine the three training results

y_pred_final = y_pred1.copy()
for i in range(len(y_pred1)):
    n_ones = y_pred1[i] + y_pred2[i] + y_pred3[i]
    if n_ones>1:
        y_pred_final[i] = 1
    else:
        y_pred_final[i] = 0


cl_rep = classification_report(y_test, y_pred_final)
print(cl_rep)
              precision    recall  f1-score   support

           0       0.99      0.78      0.87      7985
           1       0.36      0.93      0.52      1058

    accuracy                           0.80      9043
   macro avg       0.67      0.86      0.70      9043
weighted avg       0.91      0.80      0.83      9043
# The F1 Score for both classes is lower than oversampling and undersampling 

CONCLUSION: Oversampling generated the best F1 Score

TODO:

  • Add early stopping for training
  • Explore binning of some features

References:

I am grateful for the amazing content by posted by Kaggle developers. I have relied on content and tips from the following resources to develop this notebook.

https://www.kaggle.com/code/ashydv/bank-customer-clustering-k-modes-clustering

https://github.com/codebasics/deep-learning-keras-tf-tutorial/blob/master/14_imbalanced/handling_imbalanced_data.ipynb

https://www.kaggle.com/code/yufengsui/ml-project-bank-telemarketing-analysis/notebook#Part-5.-Machine-Learning:-Classification

https://www.kaggle.com/code/henriqueyamahata/bank-marketing-classification-roc-f1-recall/notebook

https://www.kaggle.com/code/gcdatkin/telemarketing-effectiveness-prediction/notebook

https://www.kaggle.com/code/volodymyrgavrysh/bank-marketing-campaigns-dataset-analysis/notebook

Releases

No releases published

Packages

No packages published