Real-Time Bitcoin Price Prediction Using Hybrid 2D-CNN LSTM Model
Saman Kazeminia, Hedieh Sajedi, Masoud Arjmand
Abstract
Due to the growing importance of the cryptocurrency market, as well as the diversity and expansion of online trading platforms, cryptocurrency technology has piqued the curiosity of a wide range of people, from market traders to researchers and analysts. Reliable price prediction is a necessity since investors face multiple challenges including market volatility, risk management, and market complexity. Therefore, numerous studies have been done using deep learning and machine learning algorithms to demonstrate their functionality and efficiency in this area. In this paper, we employed Bitcoin historical data to make predictions for the next day's closing price using a new hybrid 2D-CNNLSTM model with OPTUNA hyperparameter tuning. The dataset used to train the model was gathered using an automated web scraping technique. With the proposed model, the R2 error achieved 0.98166 and the MAPE was 0.034. Our proposed model is compared with three different models: CNN, LSTM, and GRU. The predicted results show that the proposed hybrid model is efficient for accurately predicting bitcoin prices and reliable for supporting investors to make their informed investment decisions. Additionally, the proposed model has outperformed other commonly used algorithms, namely CNN, LSTM, and GRU in terms of R2, and MAPE. This model is also capable of performing real-time forecasting.