Prediction of bitcoin stock price using feature subset optimization

Area of Science:

• Computational Finance
• Cryptocurrency Analysis
• Machine Learning Applications

Background:

• Bitcoin's increasing recognition as an investment vehicle necessitates accurate price forecasting.
• Market volatility in cryptocurrencies highlights the need for robust prediction models.
• Previous research has explored machine learning for Bitcoin price prediction, but few have investigated diverse feature subsets and modeling approaches.

Purpose of the Study:

• To identify the optimal internal feature subset for forecasting Bitcoin's price.
• To evaluate the impact of different feature categorizations on prediction accuracy.
• To enhance informed investment decisions in the volatile cryptocurrency market.

Main Methods:

• Categorization of Bitcoin's internal features into four groups: currency data, block details, mining information, and network difficulty.
• Utilizing a long short-term memory (LSTM) artificial neural network for price prediction.
• Training the LSTM model on two and a half years of historical data for each feature subset.

Main Results:

• The LSTM model achieved a mean absolute error rate of 6.38% when using the block details feature category.
• Block details features demonstrated a positive correlation with Bitcoin price and low ambiguity, contributing to improved prediction accuracy.
• Compared to other feature subsets, block details provided the most accurate Bitcoin price predictions.

Conclusions:

• The study identifies block details as the most effective feature subset for predicting Bitcoin's price using LSTM models.
• This finding offers a foundation for future research into optimizing cryptocurrency forecasting models.
• Accurate Bitcoin price prediction can aid investors in navigating market volatility.