Optimalisasi Prediksi Harga Beras di Provinsi Sumatera Utara Menggunakan Model Hybrid SARIMA-LSTM

Abstract

Rice is a staple food for the people of Indonesia, making it crucial to maintain price stability and availability. A rise in rice prices can have adverse effects as it contributes to increased inflation. Therefore, rice price forecasting is essential to support decision-making by policymakers, sellers, and consumers. This research aims to develop a website that can predict rice prices in North Sumatra province. To achieve optimal predictions, a hybrid model combining SARIMA and LSTM will be built to utilize the advantages of both models. Tests were conducted by comparing the performance of individual SARIMA and LSTM models with the Hybrid SARIMA-LSTM model. The results show that the SARIMA model tends to be less accurate in capturing the non-linear patterns of actual data. Meanwhile, the LSTM model performs better in capturing price fluctuations but still exhibits significant errors in some predictions. The Hybrid model provides the most optimal results, with an MAE 24,168 and MAPE of 0,158% for premium rice, and an MAE 21,680 and MAPE of 0,155% for medium rice.