Penerapan Model Gated Recurrent Unit-Temporal Fusion Transformer (GRU-TFT) Untuk Prediksi Harga Beras

'Azzah, Farhana Laila (2025) Penerapan Model Gated Recurrent Unit-Temporal Fusion Transformer (GRU-TFT) Untuk Prediksi Harga Beras. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5002211059-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (12MB) | Request a copy

Abstract

Beras merupakan komoditas pangan utama di Indonesia yang memiliki peran strategis dalam menjaga ketahanan pangan nasional. Konsumsi beras yang tinggi, mencapai lebih dari 100 kilogram per kapita per tahun, menjadikan stabilitas harganya sangat penting bagi kesejahteraan masyarakat dan perekonomian negara. Namun, harga beras sering mengalami fluktuasi signifikan akibat ketidakstabilan pasokan, dinamika permintaan, musim panen, serta kebijakan pemerintah. Lonjakan harga yang tidak terprediksi dapat menekan daya beli rumah tangga berpendapatan rendah dan berpotensi memicu inflasi. Kondisi ini mendorong perlunya pendekatan prediksi yang mampu memahami karakteristik data harga beras yang bersifat multivariat dan dipengaruhi faktor waktu secara dinamis. Penelitian ini menerapkan model Gated Recurrent Unit-Temporal Fusion Transformer (GRU-TFT) untuk memprediksi harga beras 14 hari ke depan dengan memanfaatkan data historis harga beras, harga gabah, stok beras, luas panen, dan informasi kalender. GRU digunakan untuk menangkap pola jangka panjang dalam data deret waktu, sedangkan TFT dengan temporal self-attention dan variable selection network memungkinkan pemilihan fitur paling relevan pada setiap periode waktu. Proses penelitian meliputi pengumpulan data, pra-pemrosesan melalui pipeline, pembentukan sliding window, serta pelatihan model menggunakan beberapa konfigurasi parameter dengan loss function DILATE dan quantile loss. Beberapa konfigurasi diuji untuk menemukan parameter terbaik, termasuk variasi dimensi representasi model, jumlah attention heads, dan jumlah epoch pelatihan. Hasil penelitian menunjukkan bahwa model GRU-TFT menghasilkan prediksi yang konsisten pada sebagian besar skenario, dengan performa yang lebih baik pada beras medium ketika menggunakan quantile loss, sedangkan penggunaan DILATE loss memberikan hasil yang lebih optimal pada beras premium.
========================================================================================================================
Rice is a major food commodity in Indonesia that plays a strategic role in maintaining national food security. High rice consumption, reaching more than 100 kilograms per capita per year, makes price stability very important for the welfare of the people and the country’s economy. However, rice prices often experience significant fluctuations due to supply instability, demand dynamics, harvest seasons, and government policies. Unpredictable price spikes can strain the purchasing power of low-income households and potentially trigger inflation. This situation underscores the need for a predictive approach capable of understanding the multivariate nature of rice price data and its dynamic time-dependent factors. This study applies the Gated Recurrent Unit-Temporal Fusion Transformer (GRU-TFT) model to predict rice prices 14 days in advance by utilizing historical data on rice prices, paddy prices, rice stocks, harvest area, and calendar information. GRU is used to capture long-term patterns in time series data, while TFT with temporal self-attention and variable selection network enables the selection of the most relevant features at each time period. The research process includes data collection, pre-processing through a pipeline, forming a sliding window, and training the model using several parameter configurations with the DILATE loss function and quantile loss. Several configurations were tested to find the optimal parameters, including variations in model representation dimensions, the number of attention heads, and the number of training epochs. The results show that the GRU-TFT model produces consistent predictions in most scenarios, with better performance for medium-grade rice when using quantile loss, while the use of DILATE loss provides more optimal results for premium rice.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	fluktuasi harga beras, GRU, ketahanan pangan, prediksi, temporal fusion transformer. rice price fluctuation, GRU, food security, prediction, temporal fusion transformer.
Subjects:	Q Science > QA Mathematics > QA276 Mathematical statistics. Time-series analysis. Failure time data analysis. Survival analysis (Biometry) Q Science > QA Mathematics > QA76.6 Computer programming. Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
Divisions:	Faculty of Mathematics and Science > Mathematics > 44201-(S1) Undergraduate Thesis
Depositing User:	Farhana Laila 'azzah
Date Deposited:	04 Aug 2025 01:58
Last Modified:	04 Aug 2025 01:58
URI:	http://repository.its.ac.id/id/eprint/126590

Actions (login required)

View Item