Predictive Commerce: Harnessing Machine Learning for E-Commerce Intelligence

Abstract

This article presents a comparison of various recommendation algorithms, including collaborative filtering, association rule mining, and machine learning approaches for product recommendations in an E-commerce setting. The dataset, consisting of 1,048,100 records with features like user ID, product ID, ratings, and timing, is used to predict ratings and recommend products. Traditional algorithms like Apriori and FP-Growth, combined with K-means clustering for anomaly detection and dashboard visualization via Power BI, provide insights into product recommendation strategies. Among the methods evaluated, SVD achieved superior performance, with an RMSE of 1.31 and MAE of 1.04. However, in our project, we employed the LightGBM algorithm, achieving an accuracy of 71%. This highlights the difference in effectiveness between matrix factorization approaches and gradient-boosting techniques in handling large-scale recommendation tasks.