Optimized Bayesian Diagnostic Assistant: Superior Accuracy with Minimal Patient History

Effective medical treatment depends on an accurate and fast diagnosis, yet many AI-driven diagnostic models suffer from ambiguity, inadequate patient histories, and interpretability issues. Using Bayesian networks, this study introduces an Optimized Bayesian Diagnostic Assistant that enhances clinical decision-making and probabilistic reasoning. Diagnostic accuracy, sparse data performance, inference speed, and model interpretability were assessed in a comparison study versus deep learning, decision trees, rule-based systems, Random Forest, Support Vector Machines (SVMs), k-Nearest Neighbors (k-NN), and Naïve Bayes. The Bayesian model outperformed deep learning (68.3% sparse data accuracy) and other models by achieving 92.5% diagnostic accuracy, 96.4% top-3 accuracy, and 87.2% accuracy with sparse data. Despite having a little longer inference time (1.2s) than deep learning (0.8s), the Bayesian technique provided more flexibility, dynamic learning, and decision-making transparency. Its resilience in managing difficult circumstances was validated by statistical analysis, such as regression and correlation tests. These results demonstrate the superiority of Bayesian networks as an AI framework for clinical diagnosis, especially in settings with incomplete patient histories. In order to facilitate scaled AI- assisted clinical decision-making, future research will concentrate on improving computing efficiency, real-time adaptation, and seamless interaction with electronic health records (EHRs).