Building an Artificial Neural Networks (ANN) to access the churn rate of a customer (whether a customer will leave the bank or not)

Suppose certain bank has seen the unusual churn rate (leaving the company) of customers and what to know why ? The task is to build the model using ann to predict if certain customer has probability to leave the bank or not.

The dataset contains the snapshot of 10000 customer details randomly selected before 6 months. The customer details contains the following features (independent features)

• CustomerID
• Surname
• CreditScore
• Geography (country they belong to (3 countries in this case))
• Gender
• Age
• Tenure
• Balance
• NumOfProducts (no. of products customer own)
• HasCrCard (if customer has credit card or not)
• IsActiveMember
• Estimated Salary

After 6 months, the bank has data if the customer leave the bank or not (dependent feature) Exited column with 1(leaved the bank) and 0(still with the bank)

Artificial Neural Network is used for creating this geo-demographic segmentation model. The following steps was taken in creating the model.

• Data Pre-processing
• Building the ANN
• Making the predictions and evaluating the model
• Testing the model with one customer sample

the confusion matrix and the classification report is shown below.

The accuracy is found to be 84% which is reasonable without any parameter tuning.

As seen in the notebook, with the information available, the model predicts the new customer will not leave the bank.

As the model shows 84% accuracy, the next time the model is trained, it can show slightly different accuracy which leads to the bias-variance tradeoff. So, k-Fold cross validation can be used to fix the variance problem by splitting the training set. Moreover, the model can be improved using he Dropout Regularization to reduce over-fitting if needed. Furthermore, parameter tuning can be used (using GridSearch) tor tuning the model.