Prediksi Harga Bitcoin Menggunakan Bidirectional Gated Recurrent Unit (Bi-GRU) Dengan Pendekatan Late Fusion Pada Data Deret Waktu Multi-Frekuensi

Siahaan, TImotius Edward (2026) Prediksi Harga Bitcoin Menggunakan Bidirectional Gated Recurrent Unit (Bi-GRU) Dengan Pendekatan Late Fusion Pada Data Deret Waktu Multi-Frekuensi. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 2043221070-Undergraduate-Thesis.pdf Restricted to Repository staff only Download (2MB) | Request a copy

Abstract

Volatilitas ekstrem pada aset digital seperti Bitcoin menuntut pengembangan model prediksi yang tidak hanya akurat tetapi juga mampu menangkap dinamika pergerakan harga secara efektif. Penelitian ini bertujuan mengevaluasi kinerja arsitektur Bidirectional Gated Recurrent Unit (Bi-GRU) dalam memprediksi harga Bitcoin menggunakan data deret waktu multi-frekuensi dengan pendekatan late fusion. Dataset penelitian mencakup variabel harian yaitu harga dan volume Bitcoin, harga emas, indeks NASDAQ serta variabel bulanan berupa indeks MSCI ACWI dan Economic Policy Uncertainty selama periode Januari 2015 hingga April 2025. Metodologi penelitian meliputi tahap pra-pemrosesan data, normalisasi, serta pembagian data latih dan uji dengan proporsi 80:20. Berdasarkan hasil hyperparameter tuning, model Bi-GRU terbaik diperoleh ketika hanya menggunakan variabel internal dengan konfigurasi 130 unit, dropout 0,4, batch size 50, dan learning rate 0,00017. Konfigurasi ini menghasilkan performa prediksi yang sangat baik dengan nilai Mean Absolute Percentage Error (MAPE) sebesar 1,79% pada data uji dan 1,29% pada periode forward testing bulan Mei hingga Juli 2025. Hasil empiris menunjukkan bahwa variabel internal memiliki pengaruh yang paling dominan dalam memodelkan pergerakan harga Bitcoin dibandingkan variabel eksternal. Temuan ini mengindikasikan bahwa pergerakan harga Bitcoin sangat responsif terhadap dinamika jangka pendek, serta memberikan kontribusi teknis dalam pengembangan model prediksi berbasis data multi-frekuensi dan wawasan praktis bagi investor dalam merumuskan strategi perdagangan jangka pendek.
================================================================================================================================
The extreme volatility of digital assets such as Bitcoin necessitates the development of predictive models that are not only accurate but also capable of effectively capturing price dynamics. This study aims to evaluate the performance of a Bidirectional Gated Recurrent Unit (Bi-GRU) architecture in forecasting Bitcoin prices using multi-frequency time series data with a late fusion approach. The dataset consists of daily variables, including Bitcoin price and volume, gold price, and the NASDAQ index, as well as monthly variables comprising the MSCI ACWI index and the Economic Policy Uncertainty index, covering the period from January 2015 to April 2025. The research methodology includes data preprocessing, normalization, and a train–test split with an 80:20 ratio. Based on hyperparameter tuning results, the best Bi-GRU model was obtained using only internal variables with a configuration of 130 units, a dropout rate of 0.4, a batch size of 50, and a learning rate of 0.00017. This configuration achieved strong predictive performance, yielding a Mean Absolute Percentage Error (MAPE) of 1.79% on the test dataset and 1.29% during the forward testing period from May to July 2025. The empirical results indicate that internal variables have a more dominant influence on modeling Bitcoin price movements than external variables. These findings suggest that Bitcoin prices are highly responsive to short-term dynamics and provide both technical contributions to multi-frequency time series forecasting and practical insights for investors in formulating short-term trading strategies.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Bitcoin, Bidirectional Gated Recurrent Unit, Data Multi-Frekuensi, Late fusion, Multi-Frequency Data
Subjects:	H Social Sciences > HA Statistics > HA30.3 Time-series analysis H Social Sciences > HB Economic Theory > Economic forecasting--Mathematical models. Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
Divisions:	Faculty of Vocational > 49501-Business Statistics
Depositing User:	Timotius Edward Siahaan
Date Deposited:	10 Jun 2026 01:22
Last Modified:	10 Jun 2026 01:22
URI:	http://repository.its.ac.id/id/eprint/133415

Actions (login required)

View Item