A novel hybrid XAI-RL framework for IoB-driven risk-adjusted portfolio optimization

In this study, we present a novel hybrid portfolio‐management framework within an Internet of Behaviors (IoB) ecosystem that brings together rule‐based heuristics, explainable AI (XAI), and reinforcement learning to make robust, risk‐aware trading decisions. Our system harnesses FinBERT to extract daily sentiment scores from social media, fusing these signals with technical indicators in an LSTM forecasting model whose hyperparameters are optimally tuned via random search. Next, we apply TIME and TimeSHAP explainability tools to gauge how much we trust each prediction, and we adjust our entry, stop-loss, and take-profit levels on the fly based on that confidence. At the same time, a Q‐learning agent learns to scale position sizes by observing recent volatility, drawdown, and explainability confidence, maximizing a risk‐aware reward that balances profit against downside exposure. When backtested on Tesla data, our XAI‐driven hybrid system delivered a 60 % total return, with a Sharpe of 2.11, Sortino of 2.98, and Calmar of 17.9—all while keeping drawdowns below 4 %. It beat both a purely rule-based strategy and a buy-and-hold approach, and the integrated CVaR checks and circuit breakers stopped extreme losses, aligning with industry risk standards. Our findings underscore how embedding XAI‐derived confidence into reinforcement‐learned risk policies can yield state‐of‐the‐art risk‐adjusted performance, enhance trust through transparency, and pave the way for behavioral‐data‐driven financial decision‐support in next‐generation IoB platforms.