Abstract:
Access to essential medicines is lacking globally, with over 50% in Africa and Asia, and accurate demand prediction is crucial for improving access. However, no studies have been conducted in the study area to evaluate the accuracy of predictive models for forecasting future essential medicine demand. The general objective of this study was to identify the most effective predictive model for forecasting essential medicine demand at public health facilities via both traditional and machine learning techniques. A cross-sectional study was conducted from September 2023-
-October 2023, using five-year consumption data from the Ethiopian Pharmaceutical supply service quantification and forecasting department. The data were analyzed in Python, and model accuracy was assessed via R-squared and root mean squared error. The consumption data were divided into training and testing datasets. The naïve, autoregressive integrated moving average, linear regression, support vector machine, random forest, artificial neural network, k-nearest neighbor, extreme gradient boosting, and gradient boosting models were used to train and test the prediction of essential medicine demand. The extreme gradient boosting model outperformed all the other models, achieving an 87% fit on the test and training datasets with a low root mean squared error. The gradient boosting, support vector machine and artificial neural network models fit 87%, 88%, and 84% of the training set, respectively, with gradient boosting and support vector machine also fitting 87% of the test set and artificial neural network fitting 83%. In contrast, the linear regression fitted 60% of the training set and 62% of the test set, with a higher root mean squared error. The autoregressive integrated moving average model exhibited moderate performance, with a 62% R2 value and a root mean squared error of 0.51, whereas the naïve model performed poorly, with a root mean squared error of 0.55 and a low R-squared value of 48%. The random forest and k-nearest neighbor models were overfitted. In conclusion, this study revealed that the extreme gradient boosting model provided comparable model fitness with lowest error estimates between the training and test datasets. It is recommended that Ethiopian Pharmaceutical supply service Gondar Hub and health facilities adopt advanced machine learning models to improve the accuracy of essential medicine demand forecasting.
