Bank customer churn prediction using neural networks is a valuable business project because it helps banks identify customers who are at risk of leaving and take proactive measures to retain them. By analyzing customer data and predicting churn, banks can gain insights into why customers leave and what factors influence their decision to stay or go. This information can be used to improve customer experience, tailor marketing and promotional activities, and develop strategies to increase customer loyalty. Ultimately, reducing churn can have a significant positive impact on a bank's bottom line, as it is typically more cost-effective to retain existing customers than to acquire new ones.
Bank Customer Churn Modelling (Kaggle)
- RowNumber: the index of the row in the data
- CustomerId: unique customer identifier
- Surname: surname
- CreditScore: credit rating
- Geography: country of residence
- Gender: gender
- Age: age
- Tenure: how many years a person has been a client of the bank
- Balance: account balance
- NumOfProducts: the number of bank products used by the client
- HasCrCard: availability of a credit card
- IsActiveMember: client activity
- EstimatedSalary: estimated salary
- Exited: the fact of the client's departure (target variable)
- Exited = 1
- Not Exited = 0
Jupyter notebook:
- Data exploring and cleansing
- Data visualization
- Looking for the best accuracy deep learning model using artificial neural networks (ANN)
- Create a graphical user interface (GUI) using Tkinter
Spyder:
- Create python functions for data visualization
- Create a web application using flask and deploy it to replit.com
Link to the web application:
- https://bank-customer-churn-prediction.jihadakbr.repl.co/
- or https://replit.com/@jihadakbr/Bank-Customer-Churn-Prediction?v=1
- Note that this web application is only active when I log in to replit.com
- The datasets contain no missing values and duplicates. Moreover, several outliers have been filtered
- The neural network model achieved an accuracy of 87.0% in correctly classifying the testing data for the bank customer churn prediction project
- This accuracy is achieved by these configurations: 1 input with 11 neurons, 1 hidden layer and dropout with 6 neurons, and 1 output with 1 neuron
- A graphical user interface (GUI) has been created using Tkinter
- Deployment of a web application using Flask on replit.com has been successful