Prediction of Drug Procurement in Healthcare Supply Chains Using Machine Learning Algorithms

The effective forecasting of the demand for drugs and their lead time is important in order to ensure that their critical supply is always available without incurring additional healthcare costs and wastage within the healthcare supply chain. In this paper, we investigate the implementation of machine learning techniques utilizing XGBoost, Random Forest, Extra Trees, and Linear Regression models for increased accuracy and flexibility of demand and lead time prediction. Sales figures, seasonal trends, and other external events serve as real-time data inputs for the proposed model which tries to respond to the changes in the demand for the drugs in an automated fashion. The models are evaluated with performance indicators which include Mean Squared Error (MSE), Mean Absolute Error (MAE), R squared, and Percentage Error, where XGBoost showed the best results. Moreover, SHAP explains the impacts of certain features so that relevant inventory control decisions can be made to further enhance management efficiency. With a prototype developed in Streamlit, the proposed solution is found to be usable and scored ‘Good’ on the System Usability Scale. This work highlights the opportunity machine learning presents to healthcare logistics and serves to inspire further work on automated adaptive forecasting systems.