Postulation of Customer Retention in Banking Sector using Machine Learning and Principal Component

Abstract

Recently, there is a rapid growth in technological improvement in banking sector. The entire world is using the banking service for managing their financial and property assets. As of now, all the technological advancements are applied to banking sector to facilitate the customers with proper operational excellence. In this view, the bank has complete responsibility in serving the people with their modern application to save their time and wealth. So the customer value analysis is needed for the bank to enrich the marketing growth and turnover of the bank. But still, the prediction of customer churn remains a challenging issue for the banking sector for analyzing the profit growth. With this view, we focus on predicting the customer churn for the banking application. This paper uses the churn modeling data set extracted from UCI Machine Learning Repository. The anaconda Navigator IDE along with Spyder is used for implementing the Python code. Our contribution is folded is folded in three ways. First, the data set is applied to various classifiers like Logistic Regression, KNN, Kernel SVM, Naive Bayes, Decision Tree, Random Forest to analyze the confusion matrix. The Performance analysis is done by comparing the metrics like Precision, Recall, FScore and Accuracy. Second the data set is subjected to dimensionality reduction method using Principal component Analysis and then fitted to the above mentioned classifiers and their performance analysis is done. Third, the performance analysis is done for the dataset by comparing the metrics with and without applying the dimensionality reduction. A Performance analysis is done with various classification algorithms and comparative study is done with the performance metric such as accuracy, precision, recall, and f-score. The implementation is carried out with python code using Anaconda Navigator. Experimental results shows that before applying dimensionality reduction PCA, the Random Forest classifier is found to be effective with the accuracy of 86%, Precision of 0.85, Recall of 0.86 and FScore of 0.84. Experimental results shows that after applying dimensionality reduction, the 2 component PCA with the kernel SVM classifier is found to be effective with the accuracy of 81%, Precision of 0.81, Recall of 0.81 and FScore of 0.74. compared to other classifiers.