Mining Customer Churns for Banking Industry using K-means and Multi-layer Perceptron

Customer churns prediction is a major concern for banking industries all over the world, and market development is happening a more considerable task, and a more important challenge is taking place in business growth. The revenue of the banking sector totally depends on its important customers. By reducing the churn customers, commercial banks have many benefits on both gaining more profits and improving core competitiveness among the competitors. In this study, a mining customer churns using K-means and Multi-Layer Perceptron (MLP) is proposed for forecasting of customer churns in the banking industry. Churn-Modelling dataset from Kaggle site is used in this study. To develop the customer churns prediction model, the dataset is first cleaned, preprocessed and then K-means is used to find similar customer groups. To improve cluster quality, Silhouette method is applied before K-means. After that, Multi-Layer Perceptron classifier is applied to predict whether each customer from each group can leave or not from a bank. The proposed model gets good accuracy with low training time in all customers groups. The results show that the proposed mining customer churns prediction is able to find a certain number of customer churns identifying customer behavior and churn happens. To compare with the proposed model, the methods of K-means + SVM and the previous study on the same dataset are used.