Optimizing Random Forest Models with Snake Optimization Algorithm for Predicting E-commerce User Purchase Behaviour

Abstract

This study proposes a Snake Optimization-based Random Forest (SO-RF) model for predicting ecommerce user behavior. Key user interaction metrics, including browsing records, purchase history, search keywords, click rate, dwell time, add-to-cart times, user comments, and visit time, serve as input features, while user conversion rate and purchase rate are the target metrics. The dataset undergoes preprocessing and feature engineering to extract meaningful patterns. The Snake Optimization (SO) algorithm fine-tunes the hyperparameters of the Random Forest (RF) model, enhancing predictive performance and generalization. Experimental results demonstrate that SO-RF outperforms conventional RF, Simulated Annealing-based RF (SA-RF), and Sparrow Search Algorithm-based RF (SSA-RF) on the test set, achieving an MAE of 0.31959, MAPE of 1.6652, MSE of 0.17625, RMSE of 0.41983, and R² of 0.96678. These findings provide valuable insights for e-commerce platforms, enabling personalized marketing strategies, improved user experience, and enhanced sales performance through accurate behavior prediction.