Providing an Evaluation Model for Medical Machine Learning in the Case of Heart Disease

Background: Cardiovascular diseases, the global number one killer, require early diagnosis to reduce premature mortality and enhance quality of life. Decision tree algorithms, whose transparency and credibility are highly valued, were used in order to capture intelligible diagnostic rules for the prediction of heart disease. They were validated and tested by doctors as clinically acceptable.Method: This study experimented on a heart disease data set with statistical tests, splitting it ۸۰:۲۰ into training and test set for distribution studies. A decision tree method generated diagnostic rules from the training set, and PPV or NPV and Support for each rule were calculated. Rules with value less than threshold were removed, and the remaining rules were tested, recalculating PPV/NPV and Support. Non-compliant rules were removed, and clinicians reviewed final rules for clinical usability.Result: This study statistically analyzed a heart disease dataset, splitting it ۸۰:۲۰ into training and test sets, with distributions validated. A decision tree algorithm generated diagnostic rules from the training set, assessed for positive predictive value (PPV) or negative predictive value (NPV) and Support. Rules below thresholds were discarded, and non-compliant adjusted rules were eliminated. Physicians evaluated the final rules for clinical acceptability.Conclusion: This article highlights the crucial role of expert-based qualitative evaluation in validating and optimizing decision tree-induced rules. Optimization rules are accepted and satisfy more than original rules, as shown through comparisons of expert ratings. The findings underscore the necessity of model accuracy, interpretability, and clinical acceptability for the implementation of AI systems in health care.Background: Cardiovascular diseases, the global number one killer, require early diagnosis to reduce premature mortality and enhance quality of life. Decision tree algorithms, whose transparency and credibility are highly valued, were used in order to capture intelligible diagnostic rules for the prediction of heart disease. They were validated and tested by doctors as clinically acceptable. Method: This study experimented on a heart disease data set with statistical tests, splitting it ۸۰:۲۰ into training and test set for distribution studies. A decision tree method generated diagnostic rules from the training set, and PPV or NPV and Support for each rule were calculated. Rules with value less than threshold were removed, and the remaining rules were tested, recalculating PPV/NPV and Support. Non-compliant rules were removed, and clinicians reviewed final rules for clinical usability. Result: This study statistically analyzed a heart disease dataset, splitting it ۸۰:۲۰ into training and test sets, with distributions validated. A decision tree algorithm generated diagnostic rules from the training set, assessed for positive predictive value (PPV) or negative predictive value (NPV) and Support. Rules below thresholds were discarded, and non-compliant adjusted rules were eliminated. Physicians evaluated the final rules for clinical acceptability. Conclusion: This article highlights the crucial role of expert-based qualitative evaluation in validating and optimizing decision tree-induced rules. Optimization rules are accepted and satisfy more than original rules, as shown through comparisons of expert ratings. The findings underscore the necessity of model accuracy, interpretability, and clinical acceptability for the implementation of AI systems in health care.

کلیدواژه ها:

Heart disease ، Decision tree ، Artificial intelligence