Demand Forecasting in the Tunisian Pharmaceutical Industry: A Comparative Study

Demand forecasting is critical for the pharmaceutical industry to ensure efficient production, inventory management, and distribution, especially in dynamic and competitive markets. This study addresses the challenges of accurate demand prediction within the Tunisian pharmaceutical sector.

The primary aim was to compare the forecasting accuracy of two methods, Holt-Winters (HW) and Multilayer Perceptron (MLP) neural networks, for three drug categories: Antiviral, Antibiotic, and Pain Relief. Additionally, the study provides actionable recommendations to enhance forecasting strategies. A 24-month (n=24) historical sales dataset (October 2020 to September 2022) from a Tunisian pharmaceutical demand forecasting company was analyzed. The analysis utilized the Holt-Winters model to incorporate seasonal adjustments and an MLP neural network to capture complex, non-linear sales patterns.

Both models were evaluated using metrics such as mean squared error (MSE) to quantify prediction accuracy. The MLP neural network consistently and significantly outperformed the Holt-Winters method, demonstrating markedly lower MSE values (e.g., 0.0206 for Antivirals, compared to 30.06 for HW) and greater adaptability to demand variability across all drug categories.

While effective for seasonal patterns, HW struggled with irregular fluctuations and complex dynamics. This study highlights the superiority of MLP neural networks for pharmaceutical demand forecasting due to their adaptability and accuracy in handling non-linear and variable data.

The findings provide a strong quantitative basis for Tunisian pharmaceutical companies to adopt advanced machine learning techniques for more reliable planning. However, to fully realize the potential of these techniques and address the limitations outlined in this research, future research should explore hybrid models and integrate extended datasets that incorporate external market dynamics and trends.