Comprehensive Machine Learning Approach for Heart Disease Diagnosis: Integrating HDTFS, Clustering and DNN

Abstract Heart disease ranks as a top cause of mortality globally, and the key to successful management lies in its timely diagnosis. Effective treatment depends on early detection. This research proposes a unique approach to detecting cardiac illness using machine learning (ML) methods, which have lately exhibited potential in this domain, combining deep neural networks (DNN) with probabilistic classification, often known as K-means clustering. The recommended strategy was evaluated using the UCI heart disease (HD) dataset. Prior to analysis, the data underwent pre-processing to manage missing values, encode categorical variables, scale them, and normalize them. An innovative technique called hybrid decision tree-based feature selection (HDTFS) is produced by merging decision trees with correlation feature selection. K-means partitioning was then used to categorize the data into groups depending on how similar they were to one another. A DNN was trained using the pre-processed data to predict the kind of heart illness. DNNs are trained using the adaptive moment optimizer (Adam optimizer), a well-known optimization method, to further refine the results. The research uncovered that the recommended strategy performed more precisely than other cutting-edge strategies. This suggests that combining HDTFS, K-means clustering, and DNN may improve the identification of heart illness and that the Adam optimizer can further improve the model's prediction capability.