LSTM-conformal forecasting-based bitcoin forecasting method for enhancing reliability.

Area of Science:

• Financial Technology
• Computational Finance
• Machine Learning

Background:

• Cryptocurrency, a novel asset class, presents significant research opportunities due to financial technology advancements.
• Bitcoin, the leading cryptocurrency, possesses substantial research value, yet its volatile nature necessitates reliable value prediction.
• Accurate and reliable prediction of Bitcoin's value is increasingly critical in financial markets.

Purpose of the Study:

• To develop and evaluate a novel method for enhancing the reliability of Bitcoin value predictions.
• To combine Long Short-term Memory (LSTM) networks with conformal prediction (CP) techniques for improved forecasting accuracy.
• To generate verifiable confidence intervals for Bitcoin price predictions.

Main Methods:

• Feature selection using the Spearman correlation coefficient method, excluding features below 0.75 and above 0.95.
• Development and training of a Long Short-term Memory (LSTM) model for Bitcoin value prediction.
• Integration of LSTM predictions into a conformal prediction framework with quantile loss and an Average Coverage Interval (ACI) predictor.

Main Results:

• The proposed LSTM-conformal prediction (LSTM-CP) model demonstrated improved reliability in forecasting Bitcoin's value.
• Confidence intervals generated by the conformal prediction model verified the reliability of the LSTM predictions.
• The Average Coverage Interval (ACI) predictor contributed to enhancing the accuracy of the prediction results.

Conclusions:

• The combination of LSTM and conformal prediction offers a robust approach for reliable cryptocurrency value forecasting.
• The LSTM-CP model effectively addresses the challenge of volatility in Bitcoin price prediction.
• This research provides a valuable tool for financial analysts and researchers seeking dependable cryptocurrency market insights.