Comparison of Triple Exponential Smoothing and Support Vector Regression Algorithms in Predicting Drug Usage at Puskesmas

Abstract

Drug management is important in managing adequate drug supplies in Puskesmas, to avoid errors in controlling existing drug stock inventory, it is necessary to predict the amount of drug usage by comparing Data Mining methods and Machine Learning methods, using the Triple Exponential Smoothing (TES) and Support Vector Regression (SVR) algorithms. Implementation is done using the Python programming language. The data used is Amlodipine 10 mg and Amoxicillin 500 mg drug data with a period of 42 months, from January 2020 - June 2023. This study aims to determine the best algorithm by comparing prediction error rate using the Mean Absolute Percentage Error (MAPE) method. Based on research that has been conducted on Amlodipine 10 mg and Amoxicillin 500 mg drugs with a division of 80% training data and 20% testing data, the Triple Exponential Smoothing algorithm with an additive model produces MAPE values of 10.36% and 17.50% respectively with the "Good" category. While Support Vector Regression algorithm, with RBF kernel, complexity 1.0, and epsilon 0.1 produces MAPE values of 10.31% and 9.38% in the "Good" and "Very Good" categories, respectively. Based on this, it can be concluded that Support Vector Regression algorithm is better at predicting than the Triple Exponential Smoothing algorithm.