A hybrid CNN-BiLSTM approach for predicting the purchase behavior of customers in e-commerce

Incomplete customer profile, especially lack of dependable income information, which is one of the primary predictors of purchasing behavior, is a major obstacle to personalized marketing in e-commerce. To eliminate this gap, this paper presents a new two-stage hybrid deep learning framework. To begin with, we suggest a distinctive feature engineering pipe, which maps and projects the heterogeneous tabular customer data into an image-like format (2D). Such a representation is inputted to a hybrid Convolutional Neural Network (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) model to precisely compute the income level of customers. The second stage involves a probabilistic model that uses the predicted level of income to create individual product recommendations. Tested on a 1,420-customer sample, the suggested model, when measured on income classification as well as followed-up purchasing recommendations, shows high accuracy of 94.58% and 95% respectively, which is much higher than a variety of models used as baselines. We have shown that predicting income using behavioral and demographic data is an effective way to build a more effective and focused e-commerce recommendation system.