Peramalan Harga Cabai Merah Besar di Provinsi Jawa Timur Berbasis Spasio-Temporal Menggunakan Graph Convolutional Network–Long Short-Term Memory

Seto, Tisso Arenggo (2026) Peramalan Harga Cabai Merah Besar di Provinsi Jawa Timur Berbasis Spasio-Temporal Menggunakan Graph Convolutional Network–Long Short-Term Memory. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5003221076-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (10MB) | Request a copy

Abstract

Harga cabai merah besar merupakan komoditas pangan strategis yang memiliki volatilitas tinggi dan berkontribusi terhadap inflasi daerah, salah satunya di Provinsi Jawa Timur. Selain dipengaruhi faktor temporal, pergerakan harga juga berpotensi dipengaruhi oleh keterkaitan spasial antarwilayah. Penelitian ini bertujuan menerapkan model Graph Convolutional Network–Long Short-Term Memory (GCN–LSTM) untuk peramalan harga cabai merah besar pada tingkat kabupaten/kota di Provinsi Jawa Timur serta mengevaluasi kontribusi informasi spasial melalui perbandingan dengan model Long Short-Term Memory (LSTM). Data yang digunakan berupa harga harian cabai merah besar tingkat konsumen pada 38 kabupaten/kota di Provinsi Jawa Timur periode 1 Januari 2021 hingga 31 Desember 2025. Struktur graph spasial dibangun menggunakan pendekatan queen contiguity dengan tambahan edge antara Kota Surabaya dan Kabupaten Bangkalan. Pemodelan dilakukan menggunakan GCN–LSTM dan LSTM dengan skema one-step-ahead forecasting. Evaluasi model dilakukan menggunakan root mean squared error (RMSE) dan mean absolute percentage error (MAPE), serta dilanjutkan dengan peramalan jangka pendek menggunakan recursive forecasting. Hasil penelitian menunjukkan bahwa model LSTM memiliki performa yang lebih baik dibandingkan GCN–LSTM. Model LSTM menghasilkan RMSE sebesar Rp1945,81/kg dan MAPE sebesar 2,60%, sedangkan GCN–LSTM menghasilkan RMSE sebesar Rp3180,1/kg dan MAPE sebesar 6,40% pada data uji. Analisis hasil prediksi menunjukkan pola temporal harga antarwilayah yang relatif homogen, sementara hasil prediksi mengindikasikan adanya oversmoothing pada model GCN–LSTM. Temuan ini menunjukkan bahwa informasi spasial yang direpresentasikan melalui graph queen contiguity belum mampu meningkatkan akurasi prediksi dibandingkan pendekatan yang hanya memanfaatkan informasi temporal. Dengan demikian, pada data dan struktur graph yang digunakan dalam penelitian ini, model GCN–LSTM belum memberikan peningkatan kinerja dibandingkan model LSTM dalam peramalan harga cabai merah besar pada tingkat kabupaten/kota di Provinsi Jawa Timur.
===================================================================================================================================
Large red chili is a strategic food commodity characterized by high price volatility and significant contribution to regional inflation. In addition to temporal factors, price movements may also be influenced by spatial relationships among regions. This study aims to implement Graph Convolutional Network–Long Short-Term Memory (GCN–LSTM) model for forecasting large red chili prices at the regency/city level in East Java Province and to evaluate the contribution of spatial information through a comparison with the Long Short-Term Memory (LSTM) model. The dataset consists of daily consumer-level large red chili prices from 38 regencies/cities in East Java Province covering the period from January 1, 2021, to December 31, 2025. The spatial graph structure was constructed using the queen contiguity approach with an additional edge connecting Surabaya City and Bangkalan Regency. Forecasting models were developed using GCN–LSTM and LSTM under a one-step-ahead forecasting scheme. Model performance was evaluated using root mean squared error (RMSE) and mean absolute percentage error (MAPE), followed by short-term forecasting using a recursive forecasting approach. The results indicate that the LSTM model outperformed the GCN–LSTM model. The LSTM model achieved an RMSE of Rp1945.81/kg and a MAPE of 2.60%, whereas the GCN–LSTM model obtained an RMSE of Rp3180.1/kg and a MAPE of 6.40%. Data analysis revealed relatively homogeneous temporal price patterns across regions, while prediction results suggested the presence of oversmoothing in the GCN–LSTM model. These findings indicate that the spatial information represented through the queen contiguity graph was not able to improve forecasting accuracy compared with an approach that relied solely on temporal information. Therefore, for the dataset and graph structure used in this study, the GCN–LSTM model did not provide performance improvements over the LSTM model in forecasting large red chili prices at the regency/city level in East Java Province.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Cabai Merah Besar, Graph Convolutional Network, Long Short-Term Memory, Peramalan Spasio-Temporal, Big Red Chili, Graph Convolutional Network, Long Short-Term Memory, Spatio-Temporal Forecasting
Subjects:	Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49201-(S1) Undergraduate Thesis
Depositing User:	Tisso Arenggo Seto
Date Deposited:	01 Aug 2026 03:53
Last Modified:	01 Aug 2026 03:53
URI:	http://repository.its.ac.id/id/eprint/141428

Actions (login required)

View Item