Prediksi Harga Beras di Jawa TimurBerdasarkan Cuaca Menggunakan Menggunakan Long Short-Term Memory (LSTM)

Mufidah, Davina (2026) Prediksi Harga Beras di Jawa TimurBerdasarkan Cuaca Menggunakan Menggunakan Long Short-Term Memory (LSTM). Other thesis, Institut Teknologi Sepuluh Nopember.

Text 2043221132-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (2MB) | Request a copy

Abstract

Beras adalah komoditas pangan utama di Indonesia yang krusial bagi ketahanan pangan dan stabilitas ekonomi, dengan fluktuasi harga yang memengaruhi daya beli masyarakat. Data BPS Mei 2025 menunjukkan penurunan harga beras medium di Jawa Timur sebesar 0,15%, namun disparitas harga antar wilayah masih tinggi, membutuhkan prediksi akurat. Jawa Timur, sebagai salah satu sentra produksi terbesar, berkontribusi 20% terhadap produksi padi nasional, namun rentan terhadap perubahan iklim seperti El Nino dan La Nina yang memengaruhi panen dan harga. Fluktuasi harga beras dipengaruhi berbagai faktor kompleks, termasuk kondisi cuaca seperti curah hujan, suhu, kelembaban yang bersifat non-linear. Penelitian ini fokus pada prediksi harga beras medium, yang paling banyak dikonsumsi masyarakat Indonesia. Model Long Short-Term Memory (LSTM) diusulkan karena kemampuannya menangani ketergantungan jangka panjang dan pola nonlinier dalam data deret waktu. Studi ini akan menganalisis kinerja LSTM menggunakan data historis harga beras dan mengintegrasikan faktor cuaca untuk meningkatkan akurasi prediksi. Data harian dari Januari 2020 hingga Desember 2024 akan digunakan, bersumber dari PIHPS Nasional dan BMKG Jawa Timur. Variabel penelitian mencakup harga beras medium, curah hujan, kecepatan angin, kelembaban, penyinaran matahari, dan temperatur. Hasil penelitian menunjukkan bahwa model LSTM dengan konfigurasi timesteps 30, dropout 0.1, batch size 16, dan 150 neurons memberikan kinerja paling optimal dengan nilai Train MAPE 0.41% dan Test MAPE 1.01%. Temuan ini membuktikan bahwa memori historis satu bulan ke belakang memegang peranan penting dalam akurasi prediksi. Berdasarkan hasil peramalan, tren harga beras sepanjang tahun 2025 diproyeksikan membentuk pola gelombang musiman yang lebih stabil dibanding tahun sebelumnya, dengan kecenderungan kenaikan pada bulan Maret dan Oktober, namun melandai menuju level terendah pada bulan Desember.
==================================================================================================================================
Rice is the primary staple food commodity in Indonesia, crucial for food security and economic stability, with price fluctuations significantly affecting public purchasing power. Data from Statistics Indonesia (BPS) in May 2025 indicated a 0.15% decline in medium-quality rice prices in East Java; however, price disparities between regions remain high, necessitating accurate prediction models. East Java, as one of the major production hubs contributing 20% to the national paddy production, is vulnerable to climate change phenomena such as El Niño and La Niña, which impact harvests and prices. Rice price, fluctuations are influenced by various complex factors, including non-linear weather conditions such as rainfall, temperature, and humidity. This study focuses on forecasting the price of medium-quality rice, the most widely consumed variety in Indonesia. The Long Short-Term Memory (LSTM) model is proposed due to its capability in handling long-term dependencies and non-linear patterns within time-series data. This study analyzes the performance of LSTM using historical rice price data and integrates weather factors to enhance prediction accuracy. Daily data from January 2020 to December 2024 was utilized, sourced from the National Strategic Food Price Information Center (PIHPS) and BMKG East Java, covering variables such as medium rice price, rainfall, wind speed, humidity, solar irradiation duration, and temperature. The results indicate that the LSTM model with a configuration of 30 timesteps, a dropout rate of 0.1, a batch size of 16, and 150 neurons yielded the most optimal performance, achieving a Train MAPE of 0.41% and a Test MAPE of 1.01%. These findings demonstrate that historical memory from the preceding month plays a crucial role in prediction accuracy. Based on the forecasting results, the rice price trend throughout 2025 is projected to form a seasonal wave pattern that is more stable compared to the previous year, with an upward tendency in March and October, before tapering off to its lowest level in December.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Beras, Long Short-Term Memory, Rice
Subjects:	H Social Sciences > HD Industries. Land use. Labor > HD47 Costs H Social Sciences > HD Industries. Land use. Labor > HD9000 Agricultural industries H Social Sciences > HF Commerce > HF5658.5 Price fluctuations T Technology > T Technology (General) > T174 Technological forecasting
Divisions:	Faculty of Vocational > 49501-Business Statistics
Depositing User:	Davina Mufidah
Date Deposited:	18 Feb 2026 00:37
Last Modified:	18 Feb 2026 00:37
URI:	http://repository.its.ac.id/id/eprint/132308

Actions (login required)

View Item