<![CDATA[Cryptocurrency markets such as Bitcoin, Ethereum, and Solana exhibit high volatility, making price forecasting difficult when relying solely on conventional technical analysis. This study aims to analyze cryptocurrency candlestick patterns by utilizing Gramian Angular Field (GAF) representations and to evaluate the performance of a hybrid deep learning model combining CNN–LSTM–Transformer to support investment decision-making. The proposed method involves processing daily historical Open, High, Low, and Close (OHLC) data from three major cryptocurrency assets: Bitcoin (BTC-USD), Ethereum (ETH-USD), and Solana (SOL-USD), covering the period from January 1, 2020, to September 30, 2024, obtained from Yahoo Finance. The time-series data were transformed into 64×64 pixel GAF images and used to train a baseline CNN model as well as a hybrid CNN–LSTM–Transformer model. Model evaluation was conducted across multiple forecasting horizons, including 1 day, 7 days, 30 days, 180 days, and 1 year, and was further complemented by real-time testing using the CoinGecko API in March 2025. The results indicate that the hybrid model achieved the best performance at different horizons for each asset: BTC-USD at the 30-day horizon with an R² of 0.971 and an SMAPE of 0.77%, ETH-USD at the 1-year horizon with an R² of 0.948 and an SMAPE of 0.81%, and SOL-USD at the 1-year horizon with an R² of 0.910 and an SMAPE of 4.72%. Real-time testing demonstrated that the model consistently captured the overall price movement trends despite high market volatility. It can be concluded that the integration of GAF representations and the hybrid CNN–LSTM–Transformer model has strong potential to enhance cryptocurrency candlestick analysis and can be utilized as a component of a Decision Support System for digital asset investment.]]>
Copyrights © 2026
