Blockchain Based Financial Forecasting for Solana Price Prediction Using Hybrid CNN BiLSTM Multi Head Attention Network

This study proposes a hybrid deep learning model combining Convolutional Neural Network (CNN), Bidirectional Long Short-Term Memory (BiLSTM), and Multi-Head Attention for forecasting Solana cryptocurrency prices. The model leverages CNN layers to extract local temporal patterns, BiLSTM to capture forward and backward dependencies in sequential data, and Multi-Head Attention to assign adaptive weights to informative time steps. Feature engineering includes stationary transformations such as multi-scale returns, volatility ratios, moving average ratios, and normalized technical indicators, which enhance the learning capacity of the model. Experimental results on historical Solana data demonstrate strong regression performance with a Mean Absolute Error of 0.7322, Root Mean Squared Error of 1.0527, Mean Absolute Percentage Error of 3.69%, and an R² of 0.9471. The predicted prices closely follow actual trends, confirming the model’s ability to capture nonlinear market dynamics, although directional accuracy remains moderate at 46.63%. These findings indicate that the proposed hybrid model provides a robust framework for accurate Solana price estimation and trend analysis, and future enhancements incorporating sentiment or regime-based features may improve short-term directional forecasting for practical trading applications.