Bitcoin price detection through deep learning algorithms

Abstract

Investor interest in cryptocurrencies has lately increased due to its inherent decentralization and openness. Accurate pricing prediction is essential for developing successful trading strategies because of the unique characteristics and volatility of cryptocurrencies. The authors of this study propose a novel framework that predicts the price of the well-known cryptocurrency Bitcoin (BTC) in order to achieve this. For consistent prediction performance in an unobserved price range, the change point detection approach is employed. Time-series data are specifically split to enable segmentation-based individual normalization. In order to forecast prices, on-chain data is also collected and utilized as an input variable. The unique records that are stored on the blockchain and are essential to cryptocurrencies are referred to as "on-chain data." Furthermore, this article proposes the use of SAM-LSTM as the prediction model, which integrates several LSTM modules for on-chain variable groups and the attention mechanism. Self-attention-based multiple long short-term memory is referred to as SAM-LSTM. Through testing with real-world BTC price data and many approach parameters, the predictive power of the proposed framework has been shown. The results are positive, with the largest MAE, RMSE, MSE, and MAPE values being 0.3462, 0.5035, 0.2536, and 1.3251, respectively.