This study aims to compare the performance of Random Forest and Support Vector Machine or SVM algorithms in predicting daily blood inventory demand across 16 combinations of blood types and components, including Packed Red Cells, Whole Blood, Thrombocyte Concentrate, and Fresh Frozen Plasma at RSUD Sekayu. The dataset comprises 1,460 daily historical transaction records spanning four years from 2022 to 2025. Model evaluation was conducted using a 10-Fold Cross-Validation method assessed by Mean Absolute Error and Root Mean Squared Error metrics, supported by scatter plot analysis. The results demonstrate that the SVM algorithm with a Radial Basis Function kernel achieves superior performance, obtaining an average MAE of 0.720 bags and RMSE of 1.094 bags, compared to Random Forest with an MAE of 0.739 bags and RMSE of 1.194 bags.
Copyrights © 2026