Optimizing Retail Sales Forecasting Through a PSO-Enhanced Ensemble Model Integrating LightGBM, XGBoost, and Deep Neural Networks

Accurate sales forecasting plays a critical role in inventory management, staffing, and promotion planning, especially in retail sectors like Rossmann stores. This study addresses the sales prediction challenge using a dataset from multiple European countries. We propose an ensemble model that integrates LightGBM, XGBoost, and Deep Neural Networks (DNN), optimized through Particle Swarm Optimization (PSO) for hyperparameter tuning. The model incorporates extensive data preprocessing steps, including outlier detection, missing value imputation, and feature selection to improve data quality. Experimental results show that our model outperforms traditional approaches in terms of Root Mean Squared Percentage Error (RMSPE), Mean Squared Error (MSE), and Precision, particularly in high-sales periods such as large promotions. The proposed solution significantly improves sales forecasting accuracy, showcasing its potential for real-world application. Future work can focus on refining feature engineering and model architecture to further enhance predictive performance.