Peramalan Harga Cabai Rawit Merah dengan GRU dan TCN Berdasarkan Hasil Penggerombolan Provinsi dengan Mahalanobis Distance-Based DTW

Abstract

Cabai rawit merah merupakan komoditas pangan strategis yang memiliki volatilitas dan disparitas harga antardaerah yang tinggi sehingga diperlukan penggerombolan berdasarkan kemiripan pola harga sebelum dilakukan peramalan. Penelitian ini bertujuan menggerombolkan 34 provinsi di Indonesia menggunakan penggerombolan hierarki berbasis Mahalanobis Distance-Based Dynamic Time Warping (MDDTW), membandingkan kinerja model Gated Recurrent Unit (GRU) dan Temporal Convolutional Network (TCN), serta melakukan peramalan harga pada setiap gerombol. Data yang digunakan berupa harga cabai rawit merah mingguan periode Januari 2023 hingga April 2026 bersumber dari Pusat Informasi Harga Pangan Strategis Nasional. Hasil penelitian menunjukkan bahwa pautan rataan menghasilkan koefisien cophenetic tertinggi sebesar 0,733 dan membentuk dua gerombol optimal berdasarkan nilai Silhouette sebesar 0,438 dan Calinski-Harabasz Index sebesar 22,105. Model GRU memberikan kinerja lebih baik dibandingkan TCN, dengan nilai MAPE, MAE, dan RMSE pada data uji masing-masing sebesar 8,513%, 5.112, dan 6.729 untuk Prototype 1, serta 5,862%, 5.292, dan 7.624 untuk Prototype 2. Hasil peramalan menunjukkan bahwa model GRU mampu mengikuti kecenderungan perubahan harga pada kedua prototype, meskipun cenderung mengestimasi nilai harga lebih rendah saat terjadi lonjakan harga.

Red bird's eye chili is a strategic food commodity in Indonesia characterized by high price volatility and substantial interregional price disparities, necessitating clustering based on price pattern similarity prior to forecasting. This study aimed to cluster the 34 provinces of Indonesia using hierarchical clustering based on Mahalanobis Distance-Based Dynamic Time Warping (MDDTW), compare the forecasting performance of Gated Recurrent Unit (GRU) and Temporal Convolutional Network (TCN) models, and forecast prices for each cluster. The study used weekly red bird's eye chili price data from January 2023 to April 2026 obtained from the National Strategic Food Price Information Center (PIHPS). The results showed that the average linkage method produced the highest cophenetic correlation coefficient of 0,733 and generated two optimal clusters based on a Silhouette coefficient of 0,438 and a Calinski-Harabasz Index of 22.105. The GRU model outperformed the TCN model, achieving testing MAPE, MAE, and RMSE values of 8,513%, 5.112, and 6.729, respectively, for Prototype 1, and 5,862%, 5.292, and 7.624, respectively, for Prototype 2. The forecasting results indicated that the GRU model was able to capture the overall price trends of both prototypes, although it tended to underestimate prices during periods of sharp price increases.