The prediction of essential medicine demand using machine learning and traditional methods on EPSS Gondar hub 2018–2022 data

Abstract

No prior research has utilized predictive models for forecasting essential medicine demand in Ethiopia. Therefore, this study aimed to identify the most accurate predictive model for predicting essential medicine demand at public health facilities via traditional and machine learning techniques in Northwest Ethiopia. A cross-sectional study was conducted from September to December 2023 using five years of consumption data from the Ethiopian pharmaceutical supply service. The top ten essential medicines were identified based on defined daily dose utilization. Demand prediction was performed in Python using nine models, naïve, ARIMA, linear regression, support vector machine, random forest, artificial neural network, k-nearest neighbor, gradient boosting, and extreme gradient boosting. Model accuracy was evaluated using R squared, root mean squared error and mean absolute error. Extreme gradient boosting was the best-performing model, achieving an R 2 of 0.87 on both training and test sets with low RMSE and MAE values (0.18). Support vector machine and gradient boosting also performed strongly with test R 2 of 0.87, though with higher RMSE (0.45–0.46) and MAE (0.25–0.35). The artificial neural network showed moderate performance (R 2 ≈ 0.83) with RMSE of 0.51 and MAE of 0.31. The extreme gradient boosting model provided comparable model fitness with lower error estimates between the training and test datasets. Therefore, incorporating machine learning models into healthcare supply chains can significantly improve forecasting accuracy and ensure a more reliable supply of essential medicines.