An integrated deep neural learning architecture with embedded feature reduction for customer lifetime value prediction in fintech firms

Abstract

The financial technology (FinTech) sector has witnessed remarkable growth in recent years, significantly transforming the way individuals and businesses enable and handle financial services. Recently, deep learning (DL) models have been beneficial in attaining appropriate data concerning customer lifetime value (CLV) for the banking experts. This paper presents a deep neural network (DNN) model with embedded feature selection (FS) for CLVP (DNNEFS-CLVP) approach. An embedded FS mechanism is primarily introduced to automatically identify the most informative attributes while eliminating redundant and less relevant data features. To further enhance predictive capability, a deep hybrid architecture is developed by integrating the temporal convolutional network, a variational graph autoencoder, and a graph convolutional network. This combined framework is expected to increase the accuracy and robustness of CLV prediction while assuring efficient feature utilization. The experimental result values of the DNNEFS-CLVP method accomplishes higher proportion of very low error values across all evaluation metrics, as evidenced by mean squared error (MSE) of 0.000081, mean absolute error (MAE) of 0.007126, root mean squared error (RMSE) of 0.009011, and mean absolute percentage error (MAPE) of 0.114715, indicating strong predictive performance and minimal deviation from actual CLV values.