COMPARATIVE ANALYSIS OF METHODSTIME SERIES FORECASTING FOR PREDICTION OF DRUG SALES IN PHARMACY (CASE STUDY: CHEMICAL FARMA TAKOMA PHARMACY)

Abstract

The existence of a pharmacy is very important for the community to meet the needs of medicines. Kimia Farma Pharmacy is a company engaged in the pharmaceutical sectorhealth care company and has the largest pharmacy network in Indonesia. Predicting drug sales at pharmacies is one of the priority activities in determining future sales, this aims to control stocks so that there are no excess and shortages of stock and prevent the unavailability of drugs that consumers want to buy. In this study, prediction of drug sales was carried out by comparing methodsmachine learning that isLong Short Term  Memory (LSTM) and statistical methods ieAuto Regressive Moving Average (ARIMA) using 5 types of drug data, then the prediction results will becompared to with the evaluation methodRoot Mean Square Error  (RMSE). Based on the results of the tests performed, the RMSE value of the LSTM method was superior to the RMSE value of the ARIMA method with a difference in the RMSE ratio for the LSTM and ARIMA models for the Acitral Suspension drug which was 4.37, for Paracetamol Syrup with a value of 38.93, for Omepros with a difference in RMSE ratio of 13.60 , for Calcium D Redoxon with a RMSE difference of 1.25, and Noza Tab@100 with a RMSE difference of 11.15. Although the LSTM model produces lower RMSE results compared to the RMSE in the ARIMA model, the LSTM model that has been created is not recommended to be used because it experiences overfitting, this is because the LSTM model cannot predict accurately for data testing or for new data types.