Retail Sales Forecasting Under Demand Volatility Using a Hybrid Fuzzy Logic and XGBoost Framework

Retail Sales Forecasting Under Demand Volatility Using a Hybrid Fuzzy Logic and XGBoost Framework

Pooya OLia Lavasani1 Goudarz Saadati-Moghadam2

1) Master’s Student in Information Technology Engineering – E-Commerce
2) Assistant Professor, Department of Information Technology Engineering – E-Commerce, Islamic Revolution University

IOI Code : ABDE-ABKFKE
Publication : 6th International Conference on Advanced Research in Management And Humanities(huconf.com)
Abstract :
    Accurate sales forecasting in the retail industry faces complex challenges due to the nonlinear nature and high volatility of demand, particularly during promotional periods and pre-holiday surges. Classical and baseline forecasting models often lack satisfactory accuracy and efficiency when faced with sudden demand shocks and changes in market behavior. The aim of this research is to present a novel hybrid framework entitled “Fuzzy-XGBoost,” in which fuzzy membership functions are integrated into the structure of the XGBoost algorithm with the aim of modeling pre-holiday demand surges (Pre-Holiday Surge) and fluctuations resulting from demand shocks. The performance of the proposed model was evaluated using the real-world Rossmann dataset and compared with XGBoost, Random Forest, Support Vector Regression (SVR), and the classical ARIMA model under different scenarios. The results showed that the Fuzzy-XGBoost model, achieving an RMSE value of 899.13 and a coefficient of determination (R²) of 0.8949 at the overall level, performed better than the other models under investigation. Furthermore, in the critical pre-holiday surge scenario, the proposed model was able to reduce the RMSE from approximately 1411 for the baseline XGBoost model and 1424 for the Random Forest model to 1176.93. Finally, the results of the Diebold-Mariano Test, with a p-value of less than 0.05, confirmed the statistical significance of the improvements achieved. The findings of this research indicate that the Fuzzy-XGBoost framework can provide an effective approach for sales forecasting and managing complex demand fluctuations, particularly during periods of demand surges.

Keywords : retail sales forecasting fuzzy logic xgboost algorithm demand volatility diebold mariano test hybrid machine learning