Hybrid Attention based GRU BiLSTM (GRBiLSTM) for Banking Customer Churn Prediction

Abstract

Early Customer churn detection is a vital aspect of Customer Relationship Management. The behavioral RFM features of the customer can be used for Banking Customer Churn prediction. The transactions made by the Customer are time series dataset. The recent activity features from the transactional data has influence in predicting the churned customers. The proposed Hybrid Attention based GRU BiLSTM (GRBiLSTM) model uses an attention layer based on recent activity behavioral pattern with the BiLSTM. This model outperforms other existing models like LSTM, CNN, and BiLSTM.