Abstract:

The underlying decentralisation and transparency of cryptocurrencies has recently sparked a lot of attention from investors. Given the volatility and distinctive qualities of cryptocurrencies, precise price prediction is crucial for creating effective trading strategies. In order to do this, the authors of this paper suggest a cutting-edge framework that forecasts the price of Bitcoin (BTC), a wellknown cryptocurrency. The change point detection approach is used for steady prediction performance in unobserved price range. Timeseries data are segmented in particular so that normalisation may be carried out individually based on segmentation. On-chain data is also gathered and used as an input variable to forecast prices. On-chain data refers to the distinct records recorded on the blockchain that are intrinsic in cryptocurrencies. Additionally, this paper suggests using SAM-LSTM, which combines multiple LSTM modules for on-chain variable groups and the attention mechanism, as the prediction model. SAM-LSTM stands for self-attention-based multiple long short-term memory. The usefulness of the suggested framework in predicting BTC prices has been demonstrated in experiments using real-world BTC price data and several technique settings. The greatest MAE, RMSE, MSE, and MAPE values were 0.3462, 0.5035, 0.2536, and 1.3251, respectively, and the findings are encouraging.

Description:

Blockchain, cryptocurrency, Bitcoin, deep learning, prediction methods, change detection algorithms.