Machine Learning and Explainable Artificial Intelligence for Early Heart Disease Prediction and Clinical Decision Support: A Systematic Review

Abstract

Healthcare is a major concern, among which heart disease is considered as one of the most important diseases where community has a big concern, when it is matter of accounting the ratio of global mortality rate along with morbidity. As the people are more aware now, and there is also easy availability of data sets, there is possibility of the acceptance of machine learning (ML) methods that make improvements in computational intelligence and have enhanced complete diagnosis of cardiovascular disease prediction and solution. At the current time, the beginning of Explainable Artificial Intelligence (XAI) has solved a problem of limitation that conventional system has of black-box modelling by permitting transparency and interpretability in system that make clinical decision power strong. As healthcare applications demand both predictive accuracy and trustworthiness, the integration of ML and XAI has become an important area of research in intelligent cardiovascular care. This systematic review examines contemporary machine learning methods and explainable AI procedures engaged prediction of heart diseases and classification. By collecting databases from various scientific sources and also ensuing a well-structured examination and while keeping in mind following also all the protocol, deep study is performed using these datasets. Also, all the strategies that must be followed during pre-processing step also prepared along with the approaches that will be applied during feature selection. All the required classification algorithms will also be identified with the calculation for metrics and work on explainability methods will be performed in order to identify the gap in the previous work. Also, emphasis will be also put on use of traditional machine learning methods, deep learning approaches and ensemble learning methods, along with post-hoc explanation that are human-understandable for understanding complex AI models.  Some examples of these explainability models are SHAP, Saliency Maps, LIME, Integrated Gradients, attention-based interpretability mechanisms, Grad-CAM etc. This study also reviews and work on strengths, restrictions, gaps and practical consequences of already researched applications that present in real-world for medical diagnosis for better healthcare environments. But this becomes now mandatory to identify main research challenges that researchers are facing due to heterogeneity nature of data, privacy and security issues, imbalance of class, model generalizability, adoption by clinical practitioners, interpretability-performance trade-offs etc. All knowledge can be only gained after studying old research papers and do findings, so this research paper talks about all new developing trends and discuss future probable and research directions for emerging system with more transparency, better reliability, that must focus to benefit patients for the prediction of cardiovascular systems for medical diagnosis. The main aim in this paper is to give platform to the researchers, clinical practitioners and healthcare professionals for knowing the current scenario of cardiovascular disease prediction using machine learning approaches and also by applying explainability AI heart disease prediction models that will ease the process and also make the upcoming system with better progress and more trustworthy, operative and unfailing clinical solutions that makes system decision support one.