Please use this identifier to cite or link to this item:
http://repository.ipb.ac.id/handle/123456789/171753| Title: | Aplikasi Temporal Convolutional Network dalam Peramalan Harga Beras Harian di Jawa Barat dengan Pendekatan Multivariate Time Series |
| Other Titles: | Application of Temporal Convolutional Network on Daily Rice Price Forecasting in West Java Using a Multivariate Time Series Approach |
| Authors: | Sumertajaya, I Made Erfiani PASARIBU, REYND HAMONANGAN |
| Issue Date: | 2025 |
| Publisher: | IPB University |
| Abstract: | Harga beras sebagai komoditas strategis di Indonesia mengalami fluktuasi
yang berdampak pada kestabilan ekonomi dan kesejahteraan masyarakat. Adanya
fluktuasi, mengakibatkan diperlukannya model peramalan yang mampu
menangkap pola kompleks dan nonlinier pada data harga beras. Penelitian ini
bertujuan untuk membangun dan mengevaluasi model Temporal Convolutional
Network (TCN) multivariat untuk memprediksi harga beras di Provinsi Jawa Barat
serta melakukan peramalan untuk 30 periode mendatang. Data yang digunakan
berupa deret waktu harga beras harian dari 3 Januari 2022 hingga 30 April 2025,
dengan variabel tambahan berupa dummy bulan pada salah satu model. Ada dua
model TCN yang dibangun, masing-masing dengan peubah dan tanpa peubah
dummy bulan, kemudian dilakukan hyperparameter tuning terhadap kernel size,
dilation rate, dropout rate, dan learning rate. Tahap hyperparameter tuning disertai
dengan sliding window validation. Hasil menunjukkan bahwa model TCN tanpa
peubah dummy bulan memberikan performa terbaik dengan nilai RMSE 245,61 dan
MAPE 1,640%, lebih baik dibandingkan model dengan dummy bulan yang
memiliki RMSE 353,95 dan MAPE 2,150%. Peramalan 30 periode ke depan
menunjukkan tren harga yang menurun secara konstan. Kemampuan TCN dalam
mempelajari pola temporal yang kompleks dan hubungan non-linear menjadikan
model ini potensial sebagai alat prediksi yang baik dan dapat dipertimbangkan.
Kata kunci: dummy bulan, harga beras harian, peramalan, sliding window, TCN As a strategic commodity in Indonesia, rice prices experience fluctuations that affect economic stability and community welfare. Therefore, a forecasting model capable of capturing complex and nonlinear patterns in rice price data is required. This study aims to develop and evaluate a multivariate Temporal Convolutional Network (TCN) model to predict rice prices in West Java Province and to forecast prices for the next 30 periods. The data used consist of daily rice price time series from January 3, 2022, to April 30, 2025, with an additional monthly dummy variable included in one of the models. Two TCN models were constructed, one with and one without the dummy variable. Hyperparameter tuning was performed on the kernel size, dilation rate, dropout rate, and learning rate. The hyperparameter tuning stage was accompanied by sliding window validation. The results show that the TCN model without the dummy variable achieved the best performance, with an RMSE of 245.61 and a MAPE of 1.640%, outperforming the model with the dummy variable, which had an RMSE of 353.95 and a MAPE of 2.150%. The 30-period-ahead forecast indicates a consistently decreasing trend in rice prices. The TCN’s ability to learn complex temporal patterns and nonlinear relationships makes it a promising predictive tool that can be further considered for modeling rice price dynamics. Keywords: daily rice price, forecast, monthly dummy, sliding window, TCN. |
| URI: | http://repository.ipb.ac.id/handle/123456789/171753 |
| Appears in Collections: | UT - Statistics and Data Sciences |
Files in This Item:
| File | Description | Size | Format | |
|---|---|---|---|---|
| cover_G1401211013_fb8d108aa595463aa7fa0e0771ab7131.pdf | Cover | 510.04 kB | Adobe PDF | View/Open |
| fulltext_G1401211013_a59d315d534a43b6829976be5f924d2e.pdf Restricted Access | Fulltext | 1.29 MB | Adobe PDF | View/Open |
| lampiran_G1401211013_cc0ae0b44bc9454f8357df32bd597178.pdf Restricted Access | Lampiran | 760.05 kB | Adobe PDF | View/Open |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.