Customer Churn Prediction in Banking Sectors Using a Hyperparameter-Tuned Deep Learning Model

Abstract

The number of service providers is growing significantly across all industries. Customers have several options when determining where to put their hard-earned cash in the banking sector. Thus, customer retention and churn have become the banks’ main priorities. Previously, supervised machine learning (ML) classifiers have been utilized in research, but feature engineering for these classifiers is labor-intensive, resulting in an incomplete and overly specific feature selection. So, the proposed system proposes a hyperparameter-tuned deep learning (DL) model for customer churn prediction (CCP) in banking sectors. The proposed system mainly comprises ‘3’ phases: data preprocessing, data balancing, and CCP. Data preprocessing, such as data cleaning and normalization, is performed on the collected dataset. After that, the data balancing is done with the help of an improved synthetic minority oversampling technique (ISMOTE) to balance the preprocessed dataset. Finally, the CCP uses hyperparameter tuned and soft plus activation based on deep multi-layer perceptron (HTSADMLP). The proposed system is tested using a churn dataset of banking customers, and the empirical results demonstrate that the proposed work outperformed conventional methods with 97.81% accuracy.