Peningkatan Akurasi Prediksi Kebutuhan Obat BPJS PRB melalui Integrasi Analisis Diferensial dan Deep Learning

Abstract

ABSTRAK
Program Rujuk Balik (PRB) BPJS Kesehatan bertujuan menjamin keberlanjutan pengobatan pasien penyakit kronis. Namun, fluktuasi kebutuhan obat sering menimbulkan permasalahan overstock dan stockout di apotek mitra BPJS. Penelitian ini bertujuan mengintegrasikan analisis diferensial dan algoritma deep learning Long Short-Term Memory (LSTM) untuk meningkatkan akurasi prediksi kebutuhan obat PRB. Data yang digunakan berupa transaksi penjualan obat pasien BPJS PRB di Apotek Kimia Farma Cendrawasih periode Januari 2022 hingga Juli 2024. Analisis diferensial digunakan untuk menghitung perubahan tingkat pertama (delta 1) dan tingkat kedua (delta 2) penjualan, yang selanjutnya dijadikan fitur tambahan pada model LSTM. Evaluasi model dilakukan menggunakan metrik Mean Squared Error (MSE), Mean Absolute Error (MAE), dan Mean Absolute Percentage Error (MAPE). Hasil penelitian menunjukkan bahwa integrasi analisis diferensial dengan LSTM mampu meningkatkan akurasi prediksi, dengan model terbaik menghasilkan nilai MAE rata-rata di bawah 20 untuk sebagian besar produk. Temuan ini berimplikasi pada peningkatan efektivitas perencanaan dan pengadaan obat PRB berbasis data historis dan tren perubahan.
Kata Kunci: 
Prediksi Obat, BPJS PRB, LSTM, Deep Learning, Analisis Diferensial

ABSTRACT
The BPJS Kesehatan Rujuk Balik Program (PRB) aims to ensure the continuity of treatment for patients with chronic diseases. However, fluctuations in medicine demand frequently cause overstock and stockout problems at BPJS partner pharmacies. This study aims to integrate differential analysis and the Long Short-Term Memory (LSTM) deep learning algorithm to improve the accuracy of PRB medicine demand forecasting. The data used consist of transaction records of PRB patient medicine sales at Kimia Farma Cendrawasih Pharmacy from January 2022 to July 2024. Differential analysis was applied to calculate the first-order change (delta 1) and second-order change (delta 2) in sales, which were subsequently incorporated as additional features in the LSTM model. Model performance was evaluated using Mean Squared Error (MSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE). The results indicate that integrating differential analysis with LSTM improves prediction accuracy, with the best-performing model achieving average MAE values below 20 for most products. These findings have important implications for enhancing data-driven planning and procurement of PRB medicines based on historical trends and demand dynamics.
Keyworsds: 
Medicine Forecasting, BPJS PRB, LSTM, Deep Learning, Differential Analysis