Electronic Theses and Dissertation

Fluktuasi harga komoditas pangan utama di kota banda aceh, lhokseumawe, dan meulaboh berpotensi mengganggu stabilitas ekonomi dan ketahanan pangan daerah. penelitian ini bertujuan mengidentifikasi hubungan kausalitas dan arah pengaruh harga komoditas pangan antarwilayah menggunakan uji kausalitas granger, membandingkan akurasi model extreme gradient boosting (xgboost) dan long short-term memory (lstm), serta meramalkan harga berdasarkan model terbaik. penelitian menggunakan data sekunder berupa harga mingguan sepuluh komoditas pangan utama yang diperoleh dari pusat informasi harga pangan strategis (pihps) nasional bank indonesia periode januari 2018 hingga mei 2026. kinerja model dievaluasi menggunakan mean absolute error (mae) dan symmetric mean absolute percentage error (smape). hasil uji kausalitas granger menunjukkan 49 dari 60 hubungan kausal signifikan pada taraf 5%, yang mengindikasikan adanya integrasi pasar antardaerah dengan pola berbeda pada setiap komoditas. bawang merah, minyak goreng, dan gula pasir menunjukkan hubungan dua arah penuh di seluruh pasangan wilayah, sedangkan komoditas lain menunjukkan kombinasi hubungan dua arah dan satu arah. lhokseumawe dan banda aceh bergantian berperan sebagai price leader, sedangkan meulaboh sebagai price follower, kecuali pada cabai rawit. hasil perbandingan menunjukkan lstm lebih akurat pada sembilan dari sepuluh komoditas dengan nilai smape 0,52%–14,36%, lebih rendah dibandingkan xgboost sebesar 1,89%–21,67%. pengecualian terdapat pada bawang merah di banda aceh dan lhokseumawe yang lebih akurat diprediksi menggunakan xgboost. akurasi tertinggi diperoleh pada beras, gula pasir, dan minyak goreng, sedangkan terendah pada cabai merah dan cabai rawit. peramalan selama 31 minggu menunjukkan tren yang bervariasi antarwilayah dan antarkomoditas, namun harga diproyeksikan tetap berada dalam rentang historis yang wajar tanpa fluktuasi ekstrem. kata kunci: kausalitas granger, lstm, xgboost, peramalan harga, komoditas pangan

Fluctuations in the prices of major food commodities in Banda Aceh, Lhokseumawe, and Meulaboh have the potential to disrupt regional economic stability and food security. This study aims to identify the causal relationships and directions of influence among food commodity prices across regions using the Granger causality test, compare the forecasting performance of the Extreme Gradient Boosting (XGBoost) and Long Short-Term Memory (LSTM) models, and forecast commodity prices for the following months using the best-performing model. The study employs secondary data consisting of weekly prices of ten major food commodities obtained from the National Strategic Food Price Information Center (PIHPS Nasional), Bank Indonesia, covering the period from January 2018 to May 2026. Model performance was evaluated using the Mean Absolute Error (MAE) and Symmetric Mean Absolute Percentage Error (SMAPE). The Granger causality test results reveal that 49 out of 60 tested causal relationships are statistically significant at the 5% significance level, indicating the presence of interregional market integration with varying patterns across commodities. Shallots, cooking oil, and sugar exhibit fully bidirectional causal relationships across all regional pairs, while the remaining commodities display a combination of bidirectional and unidirectional relationships. Banda Aceh and Lhokseumawe alternately serve as price leaders, whereas Meulaboh acts as a price follower, except in the case of bird's eye chili. The model comparison shows that LSTM outperforms XGBoost for nine of the ten commodities, achieving SMAPE values ranging from 0.52% to 14.36%, compared with 1.89% to 21.67% for XGBoost. The only exception is shallots in Banda Aceh and Lhokseumawe, for which XGBoost provides more accurate forecasts. The highest forecasting accuracy is achieved for rice, sugar, and cooking oil, while the lowest accuracy is observed for red chili and bird's eye chili. Forecasts for the next 31 weeks indicate varying price trends across regions and commodities; however, prices are generally projected to remain within their historical ranges without significant fluctuations. Keywords: Granger Causality, LSTM, XGBoost, Price Forecasting, Food Commodities