Comparative Analysis of Machine Learning Models for Pharmaceutical Demand Forecasting

Accurate pharmaceutical demand forecasting underpins supply chain resilience, production optimization, and cost efficiency. In this study, we benchmark three forecasting strategies-classical time-series models (Auto Regressive Integrated Moving Average, ARIMA; Seasonal Auto-Regressive Integrated Moving Average with Exogenous Variables, SARIMAX), advanced deep-learning architectures (Long ShortTerm Memory, LSTM; Transformer), and a hybrid ensemble (Random Forest (RF) + LSTM)-using real-world sales data from Consys Pharma and Renal Care. After standardized preprocessing and feature engineering, we evaluate each model's predictive performance (Mean Squared Error, MSE; Root Mean Square Error, RMSE; Mean Absolute Error, MAE; <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{R}^{\mathbf{2}}$</tex>) alongside its computational footprint (parameter counts, Floating-Point Operations per second, FLOPs). Our results reveal that deeplearning and hybrid ensembles more adeptly capture nonlinear and seasonal dynamics than classical methods, albeit with higher computational demands. To translate these findings into practice, we introduce an interactive dashboard that visualizes demand forecasts, confidence intervals, and “what-if” scenarios. This end-to-end framework equips stakeholders with actionable insights for precise production scheduling, optimized inventory management, and strategic resource allocation-ensuring timely medication availability while minimizing waste and costs.