Ret-DNN: Predictive Analytics in Retail - Enhanced Deep Learning Models for Customer Behavior Analysis

Abstract

In the competitive landscape of retail e-commerce, understanding and predicting customer behaviour is challenging for business success. This study introduces the Retail Deep Neural Network (Ret-DNN) model, a novel approach of advanced deep learning techniques to enhance predictive analytics in the e-commerce domain. The Ret-DNN model excels in predicting various aspects of customer behaviour, providing deep insights into shopping habits, and transaction distribution, and identifying popular products based on sales data. The proposed model also offers a detailed analysis of customer purchase frequency and transaction patterns by country, enabling a comprehensive understanding of customer engagement. By accurately predicting behaviours, the Ret-DNN model equips businesses with the tools to optimize marketing strategies, improve customer satisfaction, and drive significant growth in the retail e-commerce business. The proposed Ret-DNN model minimizes prediction errors and increases prediction precision, demonstrating performance with the lowest Validation Mean Absolute Error (MAE) of 0.2531 and Root Mean Square Error (RMSE) of 0.3575, along with high accuracy rates of 0.91, 0.90, and 0.92 for validation, test, and training phases, respectively. Overall, this novel Ret-DNN model achieves an average accuracy of 91%, highlighting its effectiveness in predicting customer behaviour in retail e-commerce. A future research direction is also presented as a concluding remark.