A Machine Learning Approach for E-Commerce Sales Forecasting and Price Optimization in the Digital Society Era

The rapid growth of e-commerce in the digital society has increased the need for intelligent systems capable of accurately predicting sales and optimizing pricing strategies. This study proposes an Ultra Ensemble Machine Learning Model that integrates XGBoost, LightGBM, Random Forest, and Multilayer Perceptron (MLP) to forecast e-commerce product sales and support data-driven decision-making. The model was developed using a comprehensive dataset containing product attributes, pricing information, and customer engagement indicators such as ratings and reviews. Experimental results demonstrate that the proposed model achieved superior predictive accuracy, with an R² value of 0.958, an RMSE of 120.76, and an MAE of 67.04, confirming its effectiveness in modeling complex nonlinear relationships among variables. The feature importance analysis identified ReviewToSalesRatio, NumReviews, and PriceRating as the most influential predictors, highlighting the critical role of customer engagement and pricing perception in shaping online purchase behavior. Additionally, correlation analysis revealed that pricing and discount-related factors strongly influence consumer demand, reinforcing the relevance of dynamic pricing mechanisms in digital markets. The findings of this study provide both theoretical and practical contributions by demonstrating the capability of ensemble learning to enhance forecasting accuracy and guide strategic pricing optimization. The proposed approach offers a scalable framework that can assist e-commerce platforms in improving sales prediction, optimizing promotional campaigns, and developing adaptive marketing strategies aligned with consumer behavior in the digital era.