Towards Sustainable Agentic Artificial Intelligence in Medical Diagnosis for Autonomous Heart Disease Prediction

Diagnosis of heart disease is a procedure that requires fast, accurate predictions and conventional fixed machine learning (ML) pipelines may not fully accommodate. Although models, such as Logistic Regression and Naïve Bayes are effective on structured clinical data however they do not have self-assessment or adaptive reasoning mechanisms. In this work, we present an agentic AI framework that turns the classical models into self-evaluating diagnostic agents managed by a performance conscious supervisory component. We trained and tested the system on the UCI Heart disease dataset over three different splits (70:30, 80:20, 85:15) to test its diagnostic stability with respect to data availability. The Logistic Regression model had the best overall (up to 85% accuracy) and F1-score performance across the experiments, while Naïve Bayes outperformed in recall for heart disease cases, indicating its beneficial use in sensitivity- oriented tasks. In each situation the supervisory agent automatically selected the superior model, demonstrating that agents can coordinate with one another even at relatively low levels of complexity. This work mediates interpretability with accountability by the construction classical models in a simple self-contained framework for a reliable transparent diagnosis without any additional computational effort. Conclusion: Agentic supervision may be the key to taking classical algorithms and developing them into scalable intelligent decision support tools for real-world cardiovascular care.