A Hybrid Linear Regression-XGBoost Model for Predicting Pharmaceutical Drug Demand

Accurate drug demand prediction is a crucial aspect of pharmaceutical supply chain management to prevent shortages or excess stock. The main challenge lies in the complex, unstable, and varying demand patterns across drug classification groups. This study develops a prediction model based on ATC (Anatomical Therapeutic Chemical) classification using a hybrid Linear Regression (LR) and Extreme Gradient Boosting (XGBoost) approach optimized with Grid Search (GS). The proposed model is compared with several baseline models, including ARIMA, General Regression Neural Network (GR_NN), Polynomial Neural Network (P_NN), and Radial Basis Function Neural Network (RBF_NN). Evaluation using RMSE and MAPE metrics across eight drug categories shows that the hybrid model outperforms others in most cases, particularly for drug groups with more consistent patterns. However, performance decreases in categories with high fluctuations, such as N02BE and R03. These findings indicate that the LR+XGB(GS) model is effective for capturing drug demand patterns and is feasible for implementation in data-driven pharmaceutical logistics decision support systems.