An Integrated ABC–VED, Machine-Learning Forecasting, and Optimization Framework for Building a Resilient Pharmaceutical Supply Chain: Evidence from Bangladesh

Abstract

In pharmaceutical supply chain management within the context of developing economies that rely on imports, such as Bangladesh, the probability of being affected by disruptions becomes greater as over 80% of APIs come from overseas. Giving equal treatment to all medicines during the prioritization process would not only be ineffective but could even lead to stockouts of essential medicines. In this study, a framework that consists of three stages of analysis - (1) ABC–VED analysis of criticality, (2) forecasting of demand for critical APIs using machine learning algorithms, and (3) MILP (mixed-integer linear programming) model formulation – was developed to assist the pharmaceutical manufacturing companies in prioritizing investments into supply chain resilience in terms of which medicines should be considered. The dataset used consisted of simulated but methodologically realistic information on 20 medicines, 18 APIs, 4 suppliers, and 60-months' observation period (January 2021 – December 2025). An integrated ABC–VED matrix helped to determine eight medicines (40% of the portfolio) as being critical and consumed 73.25% of the annual value of APIs. Six machine learning algorithms (Random Forest, XGBoost, LightGBM, Support Vector Regression, Artificial Neural Network, and CatBoost) were used to build a forecasting model for the six critical APIs and their performance was compared based on three primary evaluation metrics. The Artificial Neural Network algorithm showed the best performance in terms of minimizing errors in all the three evaluation metrics (MAE = 7.31%, MAPE = 7.43%, RMSE = 8.65%), beating the second one (XGBoost) by 9.7% -12.5%. The forecasts produced by the best algorithm are suggested to be used in the MILP model of procurement, inventory, and supplier allocation for further empirical investigation.