DefinePK

Abstract

Every sector is seeing a significant growth in the number of service providers. When deciding where to invest their money, customer in the banking sector have an abundance of options these days. Customer engagement and customer churn have thus emerged as major concerns for the majority of banks. This research proposes a unique Deep Learning (DL) with efficient Feature Selection (FS) strategy for Customer Churn (CC) Prediction (CCP) in a bank. The suggested system follows for steps to predict the CC. In step 1, the data is preprocessed, where; the missing values (MV) are filled and the dataset is normalized by Z-Score Normalization (ZSN). In step 2, the preprocessed data is passed to the next step in which K-Nearest Neighbor (KNN) technique is implemented to handle the imbalance dataset. In step 3, the optimal features are computed from the balanced dataset using Brownian Movement centered Dragonfly Optimization Algorithm (BMDOA). Finally, in step 4, the churn customers or non-churn customer classification is done by using Modified Weight and Activation centered Bidirectional Gated Recurrent Unit (MWABGRU), in which the weight is optimally tuned by BMDOA. The findings show that the suggested one outperformed state-of-the-art (SOTA) methods across all the evaluation metrics.