Information Entropy-Driven Adaptive Bayesian Model for Autonomous Decision-Making Using Reinforcement and LSH Feature Learning

Abstract

In this study, we propose an entropy-driven adaptive decision-making model designed to improve the robustness and accuracy of artificial intelligence autonomous systems under dynamic data environments. The model integrates a parameterized Bayesian structure with a reinforcement learning-based policy adjustment mechanism and a Locality-Sensitive Hashing (LSH) feature extractor. We evaluate the model using real-world datasets across three domains: medical diagnosis (Hospital Dataset-C with imbalanced binary labels), urban traffic flow (CityFlow-TR), and financial transactions (FinTech-Sim2024). Compared with a fixed Bayesian network, a deep neural model, and a basic threshold-triggered adaptation model, our system achieves a 22% improvement in diagnostic accuracy (from 0.68 to 0.90 for Disease C), 25% reduction in decision variance, and consistent performance across high-noise and largescale data. Statistical testing (t-test, p<0.05) confirms the significance of these improvements. Our findings demonstrate the effectiveness of entropy-triggered structural adjustment and adaptive policy tuning in enhancing real-time decision performance.