Ensemble-based deep learning techniques for customer churn prediction model
R. Siva Subramanian,B. Yamini,Kothandapani Sudha,S. Sivakumar
DOI: https://doi.org/10.1108/k-08-2023-1516
IF: 2.352
2024-05-21
Kybernetes
Abstract:Purpose The new customer churn prediction (CCP) utilizing deep learning is developed in this work. Initially, the data are collected from the WSDM-KKBox's churn prediction challenge dataset. Here, the time-varying data and the static data are aggregated, and then the statistic features and deep features with the aid of statistical measures and "Visual Geometry Group 16 (VGG16)", accordingly, and the features are considered as feature 1 and feature 2. Further, both features are forwarded to the weighted feature fusion phase, where the modified exploration of driving training-based optimization (ME-DTBO) is used for attaining the fused features. It is then given to the optimized and ensemble-based dilated deep learning (OEDDL) model, which is "Temporal Context Networks (DTCN), Recurrent Neural Networks (RNN), and Long-Short Term Memory (LSTM)", where the optimization is performed with the aid of ME-DTBO model. Finally, the predicted outcomes are attained and assimilated over other classical models. Design/methodology/approach The features are forwarded to the weighted feature fusion phase, where the ME-DTBO is used for attaining the fused features. It is then given to the OEDDL model, which is "DTCN, RNN, and LSTM", where the optimization is performed with the aid of the ME-DTBO model. Findings The accuracy of the implemented CCP system was raised by 54.5% of RNN, 56.3% of deep neural network (DNN), 58.1% of LSTM and 60% of RNN + DTCN + LSTM correspondingly when the learning percentage is 55. Originality/value The proposed CCP framework using the proposed ME-DTBO and OEDDL is accurate and enhances the prediction performance.
computer science, cybernetics