Gold Price Prediction Analysis Using Support Vector Regression and Random Forest Algorithms Based On BPS Data

Soraya, Hanin Fatma (2026) Gold Price Prediction Analysis Using Support Vector Regression and Random Forest Algorithms Based On BPS Data. Undergraduate thesis, UPN Veteran Jawa Timur.

	Text (COVER_22081010069) 22081010069_COVER.pdf Download (653kB)
	Text (BAB I) 22081010069_BAB I.pdf Download (88kB)
	Text (BAB II) 22081010069_BAB II.pdf Restricted to Repository staff only until 13 July 2029. Download (395kB)
	Text (BAB III) 22081010069_BAB III.pdf Restricted to Repository staff only until 13 July 2029. Download (504kB)
	Text (BAB IV) 22081010069_BAB IV.pdf Restricted to Repository staff only until 13 July 2029. Download (1MB)
	Text (BAB V) 22081010069_BAB V.pdf Download (16kB)
	Text (DAFTAR PUSTAKA) 22081010069_DAFTAR PUSTAKA.pdf Download (95kB)
	Text (LAMPIRAN) 22081010069_LAMPIRAN.pdf Restricted to Repository staff only until 13 July 2029. Download (182kB)

Abstract

Fluctuations in gold prices, influenced by various global and domestic economic factors, make gold price prediction a complex time series problem. A multivariate time series approach was applied by comparing the Support Vector Regression (SVR), Random Forest (RF), and Naive Forecast baseline models using daily data from January 2018 to April 2026 with 3,015 observations. Predictor variables used include global gold prices, the USD/IDR exchange rate, inflation, and world crude oil prices. To capture temporal patterns, feature engineering was performed using lag, moving average, volatility, and day_index. Data was divided chronologically (time-based split) with a ratio of 80:20, while hyperparameter optimization used Bayesian Optimization. Model validation on training data was performed using TimeSeriesSplit to adjust for the characteristics of the time series data and minimize the risk of data leakage. Test results showed that Support Vector Regression (SVR) provided the best performance with a MAPE of 0.06%, an RMSE of 1,825.82, and an MAE of 1,353.09. The Naive Forecast model ranked second with a MAPE of 0.23%, while Random Forest produced the lowest performance with a MAPE of 28.87% due to extrapolation failure. These results indicate that the combination of multivariate data, temporal feature engineering, Bayesian Optimization, and the Support Vector Regression algorithm can produce more accurate gold price predictions, while Random Forest is less suitable for time series data with strong long-term upward trends. Keywords:Gold Price Prediction, Support Vector Regression, RandomForest, Multivariate Time Series, Bayesian Optimization.

Item Type:

Thesis (Undergraduate)

Contributors:

Contribution	Contributors	NIDN/NIDK	Email
Thesis advisor	Nugroho, Budi	NIDK198009072021211005	budinugroho.if@upnjatim.ac.id
Thesis advisor	Aditiawan, Firza Prima	NIDK19860523 2021211 003	firzaprima.if@upnjatim.ac.id

Subjects:

Q Science > QA Mathematics > QA75 Electronic computers. Computer science

Divisions:

Faculty of Computer Science > Departemen of Informatics

Depositing User:

Hanin Fatma Soraya

Date Deposited:

13 Jul 2026 04:20

Last Modified:

13 Jul 2026 04:20

URI:

https://repository.upnjatim.ac.id/id/eprint/55238