Enhancing Worth Prediction in Cryptocurrency Utilizing Transformer Neural Community and Technical Indicators
Authors: Mohammad Ali Labbaf Khaniki, Mohammad Manthouri
Summary: This examine presents an modern strategy for predicting cryptocurrency time sequence, particularly specializing in Bitcoin, Ethereum, and Litecoin. The methodology integrates using technical indicators, a Performer neural community, and BiLSTM (Bidirectional Lengthy Brief-Time period Reminiscence) to seize temporal dynamics and extract important options from uncooked cryptocurrency information. The applying of technical indicators, such facilitates the extraction of intricate patterns, momentum, volatility, and traits. The Performer neural community, using Quick Consideration Through constructive Orthogonal Random options (FAVOR+), has demonstrated superior computational effectivity and scalability in comparison with the normal Multi-head consideration mechanism in Transformer fashions. Moreover, the mixing of BiLSTM within the feedforward community enhances the mannequin’s capability to seize temporal dynamics within the information, processing it in each ahead and backward instructions. That is significantly advantageous for time sequence information the place previous and future information factors can affect the present state. The proposed technique has been utilized to the hourly and every day timeframes of the main cryptocurrencies and its efficiency has been benchmarked in opposition to different strategies documented within the literature. The outcomes underscore the potential of the proposed technique to outperform current fashions, marking a big development within the area of cryptocurrency value prediction.
