PREDIKSI PERMINTAAN OBAT MENGGUNAKAN TIME SERIES DAN MACHINE LEARNING UNTUK SISTEM PERINGATAN DINI KETERSEDIAAN OBAT

Abstract

Effective pharmaceutical inventory management is crucial for sustaining healthcare services in hospitals. Inaccurate demand planning can lead to stockouts or overstock situations, resulting in budget inefficiencies of 25–40%. This study aims to develop and compare pharmaceutical demand forecasting models and implement the results into the SmartMed application, which integrates Safety Stock, Reorder Point (ROP), and an early warning system at RSUD Kota Bogor. The study employs the Knowledge Discovery in Databases methodology using data from January 2020 to December 2025. The study compared five forecasting methods—ARIMA, Exponential Smoothing, Prophet, Random Forest, and XGBoost—using MAE, RMSE, and MAPE metrics. Evaluation results indicate that Random Forest was the best model for four medication types, XGBoost for three, ARIMA for two, and Exponential Smoothing for one, while Prophet did not emerge as the best model for any. For Amlodipine 10 mg, Random Forest yielded an MAE of 715.94, an RMSE of 845.07, and a MAPE of 6.72% (rated "Very Good"). The best-performing models were subsequently incorporated into SmartMed to support inventory monitoring via Safety Stock, ROP, and the early warning system. This system is expected to help management make proactive decisions about pharmaceutical inventory.