Prediksi Harga Beras Di Jawa Timur Menggunakan Model Ensemble GRU – SVR Dengan Implementasi GUI

Dandi, Nur Faizi (2026) Prediksi Harga Beras Di Jawa Timur Menggunakan Model Ensemble GRU – SVR Dengan Implementasi GUI. Undergraduate thesis, UPN Veteran Jawa Timur.

	Text (Cover) Cover_Repo_removed.pdf Download (2MB)
	Text (Bab 1) Bab1_repo.pdf Download (553kB)
	Text (Bab 2) Bab2_repo.pdf Restricted to Repository staff only until 26 January 2029. Download (518kB)
	Text (Bab 3) Bab3_repo.pdf Restricted to Repository staff only until 26 January 2029. Download (800kB)
	Text (Bab 4) Bab4_repo.pdf Restricted to Repository staff only until 26 January 2029. Download (3MB)
	Text (Bab 5) Bab5_repo.pdf Download (243kB)
	Text (Daftar Pustaka) Dapus_repo.pdf Download (189kB)
	Text (Lampiran) Lampiran_repo.pdf Restricted to Repository staff only until 26 January 2029. Download (440kB)

Abstract

Rice price fluctuations represent a strategic challenge for economic stability and food security in East Java. Accurate price forecasting is therefore essential to support decision-making by governments, market participants, and the public, particularly in efforts to maintain price stability and ensure food availability. This study develops a rice price forecasting model based on an ensemble GRU–SVR architecture, which combines the capability of Gated Recurrent Units (GRU) to capture temporal patterns in time-series data with the ability of Support Vector Regression (SVR) to perform nonlinear regression on statistical features. The dataset consists of medium rice prices and daily weather variables (Tavg, RH_avg, RR, and ss) over the period 2021–2025. The preprocessing stage applies a 7-day sliding window, producing two types of features: (1) sequential features with dimensions (7 × 5) for the GRU model, and (2) 25 statistical features for the SVR model as part of the Enhanced Feature Engineering (EFE) framework. The GRU architecture comprises two stacked layers (128 units → 64 units) followed by a 32-unit dense layer functioning as a feature extractor. The SVR component employs a Radial Basis Function (RBF) kernel and is optimized using grid search over 192 hyperparameter combinations, covering ranges of C (100–10,000), gamma (0.0001–0.1), and epsilon (0.01–0.1). The experimental results demonstrate highly accurate predictive performance, with the best ensemble GRU–SVR configuration achieved at (C = 10000, γ = 0.0001), yielding a MAPE of 1,462%. The proposed model is further integrated into a Streamlit-based Graphical User Interface (GUI), enabling users to upload data, perform preprocessing, train the model, evaluate performance, and generate interactive 30-day-ahead rice price forecasts. Overall, this study contributes a precise, stable, and practical rice price forecasting model, offering a solution that bridges computational research and real-world applications in the field of food security.

Item Type:

Thesis (Undergraduate)

Contributors:

Contribution	Contributors	NIDN/NIDK	Email
Thesis advisor	Trimono, Trimono	NIDN0008099501	trimono.stat@upnjatim.ac.id
Thesis advisor	Wahyu Syaifullah, Jauharis Saputra	NIDN0725088601	wahyu.s.j.saputra.if@upnjatim.ac.id

Subjects:

Q Science > QA Mathematics > QA76.6 Computer Programming

Divisions:

Faculty of Computer Science > Departemen of Data Science

Depositing User:

Dandi Nur Faizi

Date Deposited:

27 Jan 2026 06:11

Last Modified:

27 Jan 2026 06:23

URI:

https://repository.upnjatim.ac.id/id/eprint/49139