Peramalan Harga Daging Ayam di Jawa Barat Menggunakan Algoritma Prophet dengan Moving Holiday Effect

Abstract

Harga daging ayam di Jawa Barat bersifat sangat fluktuatif akibat guncangan permintaan pada momen Hari Besar Keagamaan Nasional, namun metode peramalan tradisional seperti ARIMA memiliki keterbatasan dalam menangkap efek hari libur bergerak (moving holiday effect) karena dominan menangkap pola linier. Penelitian ini bertujuan mengembangkan model peramalan optimal menggunakan algoritma prophet untuk memprediksi harga daging ayam di Jawa Barat dengan mengidentifikasi kontribusi setiap komponen dekomposisi. Data harga harian periode 2020 - 2025 dianalisis menggunakan dekomposisi tren linear, komponen musiman, serta pemodelan moving holiday effect untuk Idul Fitri dan Idul Adha. Optimasi model dilakukan melalui grid search hyperparameter yang dikombinasikan dengan time series cross-validation terhadap 96 kombinasi parameter. Konfigurasi terbaik menggunakan seasonality mode additive dengan changepoint prior scale = 0.001, seasonality prior scale = 0.01, dan holidays prior scale = 1.00, menghasilkan CV-MAPE sebesar 5.739%. Model final memperoleh MAPE data latih sebesar 4,19% dan MAPE data uji sebesar 6.90%. Peramalan 30 hari ke depan memprediksi harga turun perlahan pada kisaran Rp37.000,00 - Rp36.000,00 per kilogram dengan MAPE sebesar 2.07%. Temuan ini menunjukkan bahwa pemodelan efek hari libur bergerak secara signifikan meningkatkan akurasi peramalan harga pangan musiman, sehingga algoritma Prophet dapat dijadikan instrumen pendukung pengambilan keputusan dalam kebijakan stabilisasi harga komoditas pangan strategis.

Chicken prices in West Java are highly volatile due to demand shocks during National Religious Holidays. Traditional forecasting methods, such as ARIMA, have limited capability in capturing moving holiday effects, as they predominantly address linear patterns. This study develops an optimal Prophet-based forecasting model for chicken prices in West Java by identifying the contribution of each decomposition component. Daily price data from 2020–2025 were analyzed using linear tren decomposition, seasonal components, and moving holiday effect modeling for Eid al-Fitr and Eid al-Adha. Model optimization combined hyperparameter grid search with time series cross-validation acros 96 parameter combinations. The best configuration used additive seasonality mode with changepoint prior scale of 0.001, seasonality prior scale of 0.01, and holidays prior scale of 1.00, yielding a CV-MAPE of 5.739%. The final model achieved a training MAPE of 4.19% and a testing MAPE of 6.90%. The 30-day forecast predicted stable prices between Rp37,000 and Rp36,000 per kilogram, with a MAPE of 2.07%. These findings indicate that modeling moving holiday effects significantly improves seasonal food price forecasting accuracy, positioning Prophet as a viable decision-support tool for food price stabilization policy.