Analisis Perbandingan XGBoost dan LightGBM dalam Prediksi Penjualan Ritel Walmart Store Sales

Abstract

Sales prediction is a crucial aspect in the retail industry for optimizing business strategies and inventory management. As a global retail company with a large-scale operation, Walmart faces significant challenges in efficiently managing its supply chain and inventory. This study conducts a comparative analysis between the Extreme Gradient Boosting (XGBoost) and Light Gradient Boosting Machine (LightGBM) algorithms in the context of retail sales prediction using the Walmart Store Sales dataset. The dataset consists of 6,436 records with 8 attributes. The research methodology implements a comprehensive machine learning approach, including data preprocessing, feature selection, dataset splitting (80:20), model training, and evaluation using standard metrics. The analysis results show that LightGBM provides superior prediction performance, with an MSE of 0.0341, MAE of 0.1120, RMSE of 0.1847, and R² of 0.9663. In comparison, XGBoost yields an MSE of 0.0408, MAE of 0.1194, RMSE of 0.2021, and R² of 0.9596. The consistent superiority of LightGBM across all evaluation metrics indicates that this algorithm is more optimal for the Walmart sales prediction case. Additionally, feature analysis shows that the variable Store contributes the most to the predictive model, while Fuel Price has a relatively minor impact. This study emphasizes that selecting the appropriate machine learning algorithm significantly affects optimal prediction outcomes, particularly in a complex, data-driven retail industry.