PENGGUNAAN TEMPORAL FUSION TRANSFORMER UNTUK PREDIKSI HARGA MINYAK GORENG DI JAWA TIMUR

Azis, Nauval Ihsani (2026) PENGGUNAAN TEMPORAL FUSION TRANSFORMER UNTUK PREDIKSI HARGA MINYAK GORENG DI JAWA TIMUR. Undergraduate thesis, UPN Veteran Jawa Timur.

	Text (Cover) 21083010121_Cover_cropped-3-21.pdf Download (1MB)
	Text (Bab 1) 21083010121_Bab 1.pdf Download (181kB)
	Text (Bab 2) 21083010121_Bab 2.pdf Restricted to Repository staff only until 25 January 2029. Download (423kB)
	Text (Bab 3) 21083010121_Bab 3.pdf Restricted to Repository staff only until 25 January 2029. Download (400kB)
	Text (Bab 4) 21083010121_Bab 4.pdf Restricted to Repository staff only until 25 January 2029. Download (1MB)
	Text (Bab 5) 21083010121_Bab 5.pdf Download (180kB)
	Text (Daftar Pustaka) 21083010121_Daftar Pustaka.pdf Download (151kB)

Abstract

Cooking oil is a strategic staple commodity in Indonesia whose price fluctuations directly affect household purchasing power, particularly in East Java Province. Price uncertainty driven by temporal dynamics and changes in crude palm oil (CPO) prices requires an accurate and adaptive forecasting system. This study develops a short-term forecasting model for packaged and bulk cooking oil prices using the Temporal Fusion Transformer (TFT), a deep learning model capable of capturing nonlinear relationships, where price changes do not always vary proportionally with time or CPO prices. Daily price data from April 2022 to May 2025 are processed as multi-group time series, in which packaged and bulk cooking oil are treated as two independent groups to allow the model to learn their distinct price characteristics. The model applies multi-horizon forecasting to simultaneously predict prices up to seven days ahead. Forecast uncertainty is represented through quantile intervals (q10–q90) generated using Quantile Loss. The combined evaluation shows good performance with MAE 0.1188, RMSE 0.1422, and MAPE 0.71%. In the group-wise evaluation, packaged cooking oil achieves lower errors (MAE 0.0429; MAPE 0.83%) compared to bulk cooking oil (MAE 0.1947; MAPE 1.19%), reflecting higher volatility in bulk prices. Forecast results indicate that the median price of packaged cooking oil ranges from IDR 18,050–18,110 per liter with wider uncertainty intervals, while bulk cooking oil ranges from IDR 16,560–16,570 per liter with narrower intervals. The model is implemented in a Streamlit-based graphical user interface (GUI) to support interactive analysis and data-driven decision making.

Item Type:

Thesis (Undergraduate)

Contributors: