Bitcoin Prices and Illiquidity Prediction using High-dimensional Features: LSTMXGBoost Approach

The term illiquidity in cryptocurrency refers to the state of a coin that cannot be easily exchanged for cash or another asset without reducing its value. Selling illiquid assets quickly can be difficult in some situations because the asset has little trading activity or interest. As a result, illiquid assets tend to have lower trading volumes and higher price volatility. This paper examines the relationship between Bitcoin ($BTC) price prediction and its illiquidity. The high-dimensional features of $BTC, including the hash rate, were collected in three different time periods. Three feature selections (FS) were used, including Filter, Wrappers, and Embedded. An approach based on the LSTM and XGBoost networks was proposed. LSTM was used to extract time series features and XGBoost was used to learn about these features. The time series features also used a wavelet transform to prevent 10 feature noise. In this approach Local Attention Mechanism (LAM) was used to select the 11 most relevant time series features. The proposed LSTMXGBoost approach was evaluated in 12 two cases: price prediction and liquidity prediction. This approach achieved MAE = 1.60 13 in predicting the price of the next day and MAE = 3.46 in predicting the illiquidity of 14 the next day. In the cross-validation of the proposed approach, the filter achieved the 15 best prediction result and the Wrapper approach achieved the best classification result. 16 In addition, the examination of the confusion matrices shows that the two tasks of price 17 prediction and liquidity prediction have no correlation.