👉Conclusion

To create a predictive analysis model on a nonlinear dataset such as the cryptocurrency market is certainly a challenging task. This study attempted to forecast the closing prices in the next 4 hours (H4), 12 hours (H12), and 1 day (D1) of the Solana cryptocurrency using the assistance of deep learning architectures. A diverse set of features including technical indicators, price data, and candlestick patterns of Solana with raw price data of Bitcoin were used as information sources to train a CNN-LSTM model for continuous prediction. A 30-day data array was provided along with 54 input features for the CNN to extract high-level features and feed them into the LSTM network. The model performed well on a large dataset with reasonable accuracy, indicating good generalization capability. The average absolute percentage error during training was found to be 2.2% on 3 years of training data, and the average absolute percentage error in testing was 3.1% on 1 year of testing data. The proposed framework was also compared to two similar studies and found to be superior. The CNN-LSTM method demonstrated significantly higher profits compared to the traditional buy-and-hold method, which incurred an 88% loss in one year, while our method yielded an additional profit of 288% including TP/SL. Quantifying the predicted values, which are the target variables in our model, can assist traders in determining trading setups, especially with derivative options instruments. A better predictive model could be obtained with more detailed data, but the inherent randomness of events worldwide will still introduce errors to the model. The main objective of this paper is to forecast market movements to maximize profits using deep learning methods. The model can be further transformed into a sophisticated trading setup where a computer actively trades based on the model's predictions.