Model Hybrid Vector Autoregressive-Neural Network (VAR-NN) Untuk Prediksi Nilai Ekspor Migas Indonesia Berdasarkan Harga Minyak Dunia dan Nilai Tukar (Kurs)

Nadhifah, Mei Dina Putri (2026) Model Hybrid Vector Autoregressive-Neural Network (VAR-NN) Untuk Prediksi Nilai Ekspor Migas Indonesia Berdasarkan Harga Minyak Dunia dan Nilai Tukar (Kurs). Undergraduate thesis, UPN Veteran Jawa Timur.

This is the latest version of this item.

	Text (Cover) 1522083010089.-cover.pdf Download (847kB)
	Text (Bab I) 1522083010089.-bab1.pdf Download (284kB)
	Text (Bab II) 1522083010089.-bab2.pdf Restricted to Repository staff only until 13 July 2029. Download (637kB)
	Text (Bab III) 1522083010089.-bab3.pdf Restricted to Repository staff only until 13 July 2029. Download (366kB)
	Text (Bab IV) 1522083010089.-bab4.pdf Restricted to Repository staff only until 13 July 2029. Download (1MB)
	Text (Bab V) 1522083010089.-bab5.pdf Download (252kB)
	Text (Daftar Pustaka) 1522083010089.-daftarpustaka.pdf Download (211kB)
	Text (Lampiran) 1522083010089.-lampiran.pdf Restricted to Repository staff only until 13 July 2029. Download (255kB)

Abstract

Changes in world oil prices and exchange rate (kurs) fluctuations have a significant impact on Indonesia's oil and gas (migas) export value, which is one of the vital sources of state revenue. The high volatility of both factors creates uncertainty in state revenue projections. Furthermore, data characteristics containing high noise and nonlinear patterns make accurate prediction of oil and gas export values difficult using conventional linear-based models. This study proposes a hybrid VAR-NN model integrated with Rolling Window Smoothing as a pre-processing technique to reduce noise and clarify data trend patterns. The VAR model captures linear relationships between variables, while the Neural Network models nonlinear patterns in VAR residuals. Data covers the period January 2000 to August 2025 sourced from BPS, World Bank, and Investing.com with a total of 296 monthly observations. The optimal VAR lag was determined through PACF analysis and AIC criteria, resulting in a VAR(13) specification, while Neural Network parameters were determined through Grid Search with Time Series Cross Validation. Model performance was evaluated using MSE, RMSE, and MAPE. Results show that the hybrid VAR-NN model with rolling window=6 achieves a MAPE of 6.76%, RMSE of 105,30, and MSE of 11.089,81 all smaller than the standalone VAR model (MAPE 6.92%, RMSE 108.36, MSE 11,742.29), proving that the hybrid approach improves prediction accuracy. This study also develops a Streamlit-based interface for interactive visualization of prediction results.

Item Type:

Thesis (Undergraduate)

Contributors:

Contribution	Contributors	NIDN/NIDK	Email
Thesis advisor	Damaliana, Aviolla Terza	NIDN0002089402	aviolla.terza.sada@upnjatim.ac.id
Thesis advisor	Wara, Shindi Shella May	NUPTK1850774675230252	shindi.shella.fasilkom@upnjatim.ac.id

Subjects:

Q Science > QA Mathematics
Q Science > QA Mathematics > QA76.6 Computer Programming
Q Science > QA Mathematics > QA76.87 Neural computers

Divisions:

Faculty of Computer Science > Departemen of Data Science

Depositing User:

Mei Dina Putri Nadhifah

Date Deposited:

13 Jul 2026 06:10

Last Modified:

13 Jul 2026 06:10

URI:

https://repository.upnjatim.ac.id/id/eprint/54999