A multimodal data-driven ensemble deep reinforcement learning model for automated cryptocurrency trading

This research proposes a deep reinforcement learning (DRL) framework for automated cryptocurrency trading that integrates sentiment analysis from social media with historical market data. Cryptocurrencies, as digital assets, are characterized by high volatility and continuous 24/7 market activity. Recently, the development of AI-powered trading bots, particularly those leveraging machine learning, has attracted considerable attention. However, the entry barrier for novice traders and investors remains high as it requires substantial effort to acquire knowledge about trading strategies, financial markets, assets, and investment principles. Machine learning techniques, and especially deep reinforcement learning, have demonstrated strong potential in addressing these challenges by offering adaptive and data-driven decision-making processes. In particular, deep learning and reinforcement learning methods have gained traction in active asset trading, often outperforming traditional benchmarks such as the Buy-and-Hold strategy. Furthermore, sentiment analysis plays a crucial role in automated trading, as platforms such as Twitter can significantly influence the price dynamics of cryptocurrency. In this study, we propose a multimodal approach to automated cryptocurrency trading by combining various DRL models with sentiment-based signals, aiming to improve trading performance and robustness in highly volatile markets.