Peramalan Log Return dan Klasifikasi Directional Bitcoin Menggunakan Bi-LSTM Berbasis Fitur Multi-Sumber

Wardana, Razpa Arya (2026) Peramalan Log Return dan Klasifikasi Directional Bitcoin Menggunakan Bi-LSTM Berbasis Fitur Multi-Sumber. Undergraduate thesis, UPN Veteran Jawa Timur.

Preview

Text (Cover)
22082010109_Cover.pdf
Download (951kB) | Preview

Preview

Text (Bab 1)
22082010109_Bab 1.pdf
Download (375kB) | Preview

Text (Bab 2)
22082010109_Bab 2.pdf
Restricted to Repository staff only until 25 May 2028.
Download (583kB)

Text (Bab 3)
22082010109_Bab 3.pdf
Restricted to Repository staff only until 25 May 2028.
Download (460kB)

Text (Bab 4)
22082010109_Bab 4.pdf
Restricted to Repository staff only until 25 May 2028.
Download (1MB)

Preview

Text (Bab 5)
22082010109_Bab 5.pdf
Download (303kB) | Preview

Preview

Text (Daftar Pustaka)
22082010109_Daftar Pustaka.pdf
Download (184kB) | Preview

Abstract

Bitcoin price movements are highly volatile, creating significant challenges in generating accurate predictions, particularly in determining the direction and magnitude of price changes. This study aims to develop a daily Bitcoin log return prediction model using a Bidirectional Long Short-Term Memory (Bi-LSTM) approach and to analyze the contribution of feature groups through an ablation study with a leave-one-group-out method. The dataset consists of daily historical Bitcoin data (OHLCV) combined with technical indicators, volatility measures, and derived log return features. The model is developed within a multi-task learning framework to simultaneously predict log return values (regression) and price movement direction (classification). Model performance is evaluated using RMSE, MAE, and MAPE based on reconstructed prices, as well as Directional Accuracy to assess directional prediction performance. The results indicate that the Bi-LSTM model achieves stable predictive performance with relatively low error rates and competitive directional accuracy, while the ablation study demonstrates that technical and volatility feature groups contribute significantly to performance improvement. Overall, the findings confirm that a log return and directional classification based forecasting approach using a Bi-LSTM architecture and systematic feature contribution analysis can enhance prediction stability and model interpretability in Bitcoin price forecasting.

Item Type:

Thesis (Undergraduate)

Contributors: