Mulyana, Mutya Febby (2026) Perbandingan Metode Bi-LSTM dan Bi-GRU dalam Memprediksi Harga Cryptocurrency dengan Estimasi Risiko Value at Risk dan Expected Shortfall. Other thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 5006221001-Undergraduate_Thesis.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
5006221001-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (9MB) | Request a copy |
Abstract
Cryptocurrency berkembang pesat sebagai salah satu instrumen investasi digital dengan potensi keuntungan besar, namun disertai risiko tinggi akibat volatilitas harga dan ketidakpastian regulasi. Perkembangan aset kripto di Indonesia semakin diperhatikan, terbukti dengan peningkatan jumlah investor dan penguatan regulasi oleh pemerintah, sehingga mendorong kebutuhan prediksi harga dan analisis risiko yang akurat. Penelitian ini bertujuan untuk memperoleh metode prediksi terbaik menggunakan algoritma deep learning, yaitu Bidirectional Long Short-Term Memory (Bi-LSTM) dan Bidirectional Gated Recurrent Unit (Bi-GRU), serta melakukan estimasi risiko melalui Value at Risk (VaR) dan Expected Shortfall (ES) untuk mendukung pengambilan keputusan investasi. Data yang digunakan merupakan harga penutupan harian dari empat aset kripto terpopuler, yaitu Bitcoin (BTC), Ethereum (ETH), Ripple (XRP), dan Binance Coin (BNB) untuk periode September 2018 hingga Agustus 2025. Setelah dilakukan pemodelan Bi-LSTM dan Bi-GRU dengan berbagai kombinasi variabel input dan hyperparameter, hasil penelitian ini menunjukkan bahwa Bi-LSTM merupakan metode terbaik untuk ETH dan BNB dengan MAPE sebesar 2,7534% dan 1,9154%, sedangkan Bi-GRU merupakan metode terbaik untuk BTC dan XRP dengan MAPE sebesar 1,8006% dan 3,2379%. Hasil peramalan untuk periode September 2025 memperlihatkan bahwa BTC mengalami tren kenaikan harga, BNB bergerak relatif stabil dengan kecenderungan meningkat, sedangkan ETH dan XRP cenderung mengalami penurunan harga. Pengukuran risiko menunjukkan bahwa BTC memiliki tingkat risiko paling rendah, ETH dan BNB berada pada kategori risiko menengah, dan XRP memiliki risiko tertinggi terutama pada CI 99%. Hasil backtesting memvalidasi bahwa model risiko valid pada seluruh tingkat kepercayaan untuk BTC dan BNB, namun tidak valid pada ETH CI 90% dan XRP CI 99% yang mengindikasikan adanya perilaku ekor distribusi yang lebih ekstrem pada kedua aset tersebut.
======================================================================================================================================
Cryptocurrency has grown rapidly as a digital investment instrument with large profit potential, but it is also accompanied by high risk due to price volatility and regulatory uncertainty. The development of crypto assets in Indonesia is receiving increasing attention, as evidenced by the rising number of investors and the strengthening of government regulations, which encourages the need for accurate price prediction and risk analysis. This study aims to obtain the best prediction method using deep learning algorithms, namely Bidirectional Long Short Term Memory (Bi LSTM) and Bidirectional Gated Recurrent Unit (Bi GRU), as well as to conduct risk estimation through Value at Risk (VaR) and Expected Shortfall (ES) to support investment decision making. The data used consist of daily closing prices of four major crypto assets, namely Bitcoin (BTC), Ethereum (ETH), Ripple (XRP), and Binance Coin (BNB), covering the period from September 2018 to August 2025. After modeling Bi-LSTM and Bi-GRU models with various combinations of input variables and hyperparameters, the results indicate that Bi-LSTM is the best method for ETH and BNB, with MAPE values of 2,7534% and 1,9154%, respectively. Meanwhile, Bi-GRU performs better for BTC and XRP, with MAPE values of 1,8006% and 3,2379%. Forecasting results for September 2025 indicate that BTC exhibits an upward trend, BNB shows relatively stable movement with a tendency to increase, while ETH and XRP experience declining price trends. Risk estimation shows that BTC has the lowest risk level, ETH and BNB are categorized as medium-risk assets, and XRP has the highest risk especially at the 99% CI. The backtesting results validate that the risk model is valid at all confidence levels for BTC and BNB, but not valid for ETH at the 90% CI and XRP at 99% CI, indicating more extreme tail distribution behavior in these two assets.
| Item Type: | Thesis (Other) |
|---|---|
| Uncontrolled Keywords: | Bidirectional GRU, Bidirectional LSTM, cryptocurrency, Expected Shortfall, Value at Risk |
| Subjects: | Q Science > Q Science (General) > Q325.78 Back propagation Q Science > QA Mathematics > QA276 Mathematical statistics. Time-series analysis. Failure time data analysis. Survival analysis (Biometry) Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) Q Science > QA Mathematics > QA76.9.D343 Data mining. Querying (Computer science) |
| Divisions: | Faculty of Science and Data Analytics (SCIENTICS) > Actuaria > 94203-(S1) Undergraduate Thesis |
| Depositing User: | Mutya Febby Mulyana |
| Date Deposited: | 12 Jan 2026 01:16 |
| Last Modified: | 12 Jan 2026 01:16 |
| URI: | http://repository.its.ac.id/id/eprint/129443 |
Actions (login required)
 |
View Item |