Marketing_Campaign_Strategy

Trained with 37,000 datasets to determine best campaign strategy for collecting customer’s needs and overall satisfaction.

Objective: The purpose of marketing campaign is to collect customer’s needs and overall satisfaction. There are a few essential aspects of the marketing campaign namely, customer segmentation, promotional strategy, and etc. Correctly identified strategy may help to expand and grow the bank’s revenue.

credit datasource: HackerEarth HackLive: Customer Segmentation

STEP 1) DATA LOADING

df = pd.read_csv(CSV_PATH)

STEP 2) DATA INSPECTION

Segregate data in terms of Categorical and Continuous and plot data

#Continuous Data

#Categorical Data vs Target

Check for NaN values

There are many NaN values needed to be imputed or removed #last_contact_duration:311, num_contacts_in_campaign :112, days_since_prev_campaign_contact :25831

since the days_since_prev_campaign_contact NaN values is about 80% of the data, the column will be removed from features

STEP 3) DATA CLEANING

Remove days_since_prev_campaign_contact column

df= df.drop(labels='days_since_prev_campaign_contact',axis=1)
con_column.remove('days_since_prev_campaign_contact')

Label encoder all categorical data and convert alphabet to numerical values #save label encoder pickle

Fill in NaN values with Imputer

`#Use iterative imputer to fill in remaining NaN values ii = IterativeImputer() df_II = ii.fit_transform(df) df= pd.DataFrame(df_II) df.columns= column_names

#round of data to integer value for index, i in enumerate(column_names): df[i]= np.floor(df[i]).astype('int')`

STEP 4) FEATURES SELECTION

Use Logistic Regression to determine accuracy score between continuous data and categorical data

Results:

customer_age : 0.892754447498973
balance : 0.8925648560685057
day_of_month : 0.892754447498973
last_contact_duration : 0.8996429361392865
num_contacts_in_campaign : 0.892754447498973
num_contacts_prev_campaign : 0.8918064903466363

Use Cramer's V to determine accuracy score between categorical data and categorical data

Results:

job_type: accuracy is 0.13596388961197914
marital: accuracy is 0.06396495992281373
education: accuracy is 0.07207191785440015
default: accuracy is 0.018498692474409054
housing_loan: accuracy is 0.144442194297714
personal_loan: accuracy is 0.06550522151922447
communication_type: accuracy is 0.14713602417060107
month: accuracy is 0.2713965669912321
prev_campaign_outcome: accuracy is 0.3410607961880476

All continuous data has high correlation value(>0.8) with target term_deposit_subscribed #Hence is selected as Features -> X

X = df[con_column] y = df['term_deposit_subscribed']

STEP 5) DATA PRE-PROCESSING

Use Standard Scaler to fit and scale all Features(X)

Use OneHotEncoder for target since term_deposit_subscribed is categorical

ohe= OneHotEncoder(sparse=False) y=ohe.fit_transform(np.expand_dims(y,axis=-1))

Train test split data into train and test set

x_train,x_test,y_train,y_test=train_test_split(X,y,test_size=0.3, random_state=123)

STEP 6) MODEL DEVELOPMENT

Use two_layer_model using Input, Dense, Dropout Rate, BatchNormalization features

Visualising Neural Network model plot_model(model,show_shapes=True, show_layer_names=(True))

Compile model with categorical_crossentropy loss hyperparameters

model.compile(optimizer='adam',loss='categorical_crossentropy',metrics=['acc'])

Callbacks: tensorboard_callback and early_stopping_callback

STEP 7) MODEL ANALYSIS

#Model Fitting and Testing with 100 epochs and batch size= 64

hist = model.fit(x=x_train,y=y_train,batch_size=64,epochs=100, validation_data=(x_test,y_test), callbacks=[tensorboard_callback,early_stopping_callback])

STEP 8) MODEL EVALUATION

Plot history for training accuracy and validation(test) accuracy

Plot training loss and validation(test) loss

TensorBoard function for all logs data

Model evaluation me.model_evaluation(model, x_test, y_test)

[0.2656640410423279, 0.9022643566131592]

Classification Report

                  precision    recall  f1-score   support
               0       0.91      0.99      0.95      8504
               1       0.61      0.19      0.29       991
        accuracy                           0.90      9495
       macro avg       0.76      0.59      0.62      9495
    weighted avg       0.88      0.90      0.88      9495

Confussion Matrix

STEP 9) MODEL SAVING

#saving ohe model, std scaler model and Neural network model

Discussion

#The deep learning model achieved a good performance with 90% accuracy showing 'customer_age', 'balance', 'day_of_month', 'last_contact_duration', num_contacts_in_campaign', 'days_since_prev_campaign_contact', 'num_contacts_prev_campaign'has the highest correlation (above 80%) with 'term_deposit_subscribed'

Since term_deposit_subscribed data is unbalanced, The model should be improved in the future by training more dataset into the model and reducing the underfitting model.