TEMPORAL CONVOLUTIONAL NETWORK BASED ON TIME SERIES DATA ANALYSIS FOR FOOD PRICE PREDICTION IN THE EAST JAVA MODERN MARKET

Nurdiansyah, Titis Fajar (2026) TEMPORAL CONVOLUTIONAL NETWORK BASED ON TIME SERIES DATA ANALYSIS FOR FOOD PRICE PREDICTION IN THE EAST JAVA MODERN MARKET. Undergraduate thesis, UPN Veteran Jawa Timur.

	Text (COVER) 1_COVER_22081010086_merged (1)_organized.pdf Download (695kB)
	Text (BAB I) 2_BAB 1_22081010086.pdf Download (91kB)
	Text (BAB II) 3_BAB 2_22081010086.pdf Restricted to Repository staff only until 2028. Download (716kB)
	Text (BAB III) 4_BAB 3_22081010086.pdf Restricted to Repository staff only until 2028. Download (1MB)
	Text (BAB IV) 5_BAB 4_22081010086.pdf Restricted to Repository staff only until 2028. Download (4MB)
	Text (BAB V) 6_BAB 5_22081010086.pdf Download (14kB)
	Text (DAFTAR PUSTAKA) 7_DAFTAR PUSTAKA_22081010086.pdf Download (155kB)
	Text 8_LAMPIRAN_22081010086.pdf Restricted to Repository staff only Download (296kB)

Abstract

Fluctuations in strategic food commodity prices in East Java create uncertainty for various stakeholders, ranging from producers to end consumers. Historical data from Bank Indonesia's Strategic Food Price Information Center (PIHPS) have not yet been optimally utilized to develop a reliable forecasting system. This study aims to implement a Temporal Convolutional Network (TCN) model to forecast the prices of 10 strategic food commodities in modern markets across East Java, including shallots, garlic, rice, red chilies, bird's eye chilies, chicken meat, beef, granulated sugar, cooking oil, and chicken eggs, using daily PIHPS data from 2017 to 2026.The research pipeline includes exploratory data analysis (EDA), preprocessing using forward fill for data cleaning, feature engineering with 15 temporal features, Min-Max Scaling normalization, and sliding-window transformation with a 70/15/15 data split. The TCN architecture was selected for its ability to capture long-range temporal dependencies through causal and dilated convolutions while avoiding the vanishing-gradient issues commonly associated with recurrent architectures. The model was evaluated across three forecasting horizons (H+7, H+15, and H+30) using two configurations: a Baseline model and an Optimized model developed through Optuna-based Bayesian optimization with a Tree-structured Parzen Estimator (TPE) sampler and 30 trials for each combination.The results show that the Optimized TCN model consistently outperformed the Baseline model across all forecasting horizons, reducing MAPE by 37.37% for H+7, 20.53% for H+15, and 21.12% for H+30. These improvements indicate that hyperparameter optimization plays a significant role in enhancing prediction accuracy, particularly for shorter forecasting horizons. The model was subsequently deployed through a web interface built with FastAPI and React.js to support real-time inference. This deployment demonstrates the practical feasibility of the proposed system as a decision-support tool for monitoring food price trends and assisting policymakers, market regulators, and the public in anticipating future price movements. Overall, this research contributes a data-driven approach that can be extended to other commodities or regions with similar price volatility characteristics. Future work may explore the integration of external factors such as weather conditions and supply-chain disruptions to further improve prediction accuracy.

Item Type:

Thesis (Undergraduate)

Contributors:

Contribution	Contributors	NIDN/NIDK	Email
Thesis advisor	Puspaningrum, Eva Yulia	NIDN0005078908	evapuspaningrum.if@upnjatim.ac.id
Thesis advisor	Putra, Chrystia Aji	NIDN0008108605	ajiputra@upnjatim.ac.id

Subjects:

T Technology > T Technology (General)

Divisions:

Faculty of Computer Science > Departemen of Informatics

Depositing User:

Unnamed user with email 22081010086@student.upnjatim.ac.id

Date Deposited:

07 Sep 2026 01:28

Last Modified:

07 Sep 2026 01:28

URI:

https://repository.upnjatim.ac.id/id/eprint/59943