Bitcoin Price Prediction Using SVR Optimized with Bayesian Optimization

Wahyudi, Wahyudi (2026) Bitcoin Price Prediction Using SVR Optimized with Bayesian Optimization. Undergraduate thesis, UPN Veteran Jawa Timur.

[img] Text (Cover)
21081010100.-cover.pdf
Download (4MB)
[img] Text (Bab 1)
21081010100.-bab1.pdf
Download (231kB)
[img] Text (Bab 2)
21081010100.-bab2.pdf
Restricted to Repository staff only until 22 July 2028.
Download (476kB)
[img] Text (Bab 3)
21081010100.-bab3.pdf
Restricted to Repository staff only until 22 July 2028.
Download (866kB)
[img] Text (Bab 4)
21081010100.-bab4.pdf
Restricted to Repository staff only until 22 July 2028.
Download (2MB)
[img] Text (Bab 5)
21081010100.-bab5.pdf
Download (235kB)
[img] Text (Daftar pustaka)
21081010100.-daftarpustaka.pdf
Download (259kB)
[img] Text (Lampiran)
21081010100.-lampiran.pdf
Restricted to Repository staff only
Download (209kB)

Abstract

With the world's largest market capitalization, Bitcoin exhibits strong heteroscedasticity and sharp price movements. The non-linear relationship between market variables and sharp price movements make conventional methods less effective in generating accurate predictions. Using hourly historical data (OHLCV) obtained from the Binance API from November 27, 2020, to November 26, 2025, this study employed a Support Vector Regression (SVR) approach with a Radial Basis Function (RBF) kernel automatically optimized through Bayesian Optimization. Testing was conducted using a chronological, multilevel data partitioning scheme, with 64% for training data, 16% for validation data, and 20% for test data. Meanwhile, model evaluation was conducted over three prediction horizons: 1 day (24 hours), 3 days (72 hours), and 7 days (168 hours) in the future to assess the model's accuracy and overall performance. Experimental results show that the SVR model optimized with Bayesian Optimization consistently outperforms the conventional Grid Search method across all time horizons. At the 1-day horizon, Bayesian Optimization successfully reduced the Mean Absolute Percentage Error (MAPE) by 34.46% to 1.94% and increased the coefficient of determination (R2) to 0.9480 compared to Grid Search. The most significant performance improvement in explaining data variability occurred at the 3-day horizon with an 11.97% jump in R2 to 0.8904.

Item Type: Thesis (Undergraduate)
Contributors:
ContributionContributorsNIDN/NIDKEmail
Thesis advisorParlika, RizkyNIDN0718058401rizkyparlika.if@upnjatim.ac.id
Thesis advisorVia, Yisti VitaNIDN0025048602yistivia.if@upnjatim.ac.id
Subjects: Q Science > QA Mathematics > QA76.6 Computer Programming
Divisions: Faculty of Computer Science > Departemen of Informatics
Depositing User: Wahyudi - -
Date Deposited: 23 Jul 2026 02:01
Last Modified: 23 Jul 2026 02:38
URI: https://repository.upnjatim.ac.id/id/eprint/57833

Actions (login required)

View Item View Item