- This repository will have complete machine learning and deep learning based banking churn prediction ANN model which will analyze tha probablity for a customer to leave.
- The project was deployed on Google Cloud Platform as well as completely tested on Localhost.
- The main welcome screen is made in HTML5 and CSS3 with a basic and simple design.
- Here is the main screen for the Bank Churn Prediction Interface for the bank admin.
- The bank employee has to enter the details of the customer whose churn they want to analyze.
- Below is the screenshot of the input being filled by bank employee.
- The Prediction engine is built over a deep Artificial Neural Network backed with Keras.
- I have achieved an accuracy of around ~85% on both training and testing data.
- The ANN is trained over K-fold cross validation testing over 10 rounds to find if it was underfit or overfit over the data based on the variance betweent the accuracies of the 10 rotations.
- The model is Tuned over the Hyerparametes to find the best batch_size, epoch and optimizer for generating the best possible combination for best fit model.
- The API interfacing for the deplyment on Localhost is done using Flask.
- The server is run on Local system during the staging of the project.
- Older deployment was done on Google Cloud Platform
- Recently, the final deployment was done on Heloku platform and can be accessed from the link below.
- LINK: https://banking-churn-pediction
- The final prediction of the model is the percentage of churn for that customer.
- The prediction signifies the chances of the customer to leave the services of the bank which makes the bank to focus more on such such customers and try to retain them using Sales and Marketing strategies about which I have worked in this GitHub module.
