Majority voting ensembled feature selection and customized deep neural network for the enhanced clinical decision support system

Abstract

Heart disease and diabetes are global health issues that affect people worldwide. Diabetes is becoming a significant concern, and Diabetes patients have a substantially higher risk of heart disease morbidity and mortality than people without diabetes. These conditions are associated with hospitalizations and emergency room visits, which raises healthcare expenses. An important strategy to improve health care outcomes and reduce unnecessary costs is to identify and anticipate them in patients. Clinical Decision Support Systems (CDSS) assess patient data from clinical datasets to help disease prediction and enhance treatment options for heart disease and diabetes, and other disorders. According to the literature, most CDSS have used machine learning algorithms for predicting heart disease and diabetes. These algorithms performed worthily, but the accuracy of these machine learning (ML) algorithms is lacking, especially in medical data, which contains numerous complex attributes such as resting blood pressure, serum cholesterol, fasting blood sugar, and thalassemia value. This proposed work developed a majority voting ensembled feature selection (MVEFS) technique and customized deep neural network (CDNN) to develop a CDSS for heart disease and diabetes prediction. This deep neural network-based CDSS best performing than ML-based CDSS. There are several input attributes in the clinical dataset. Some attributes are not associated with disease and have negative consequences when used in clinical data analysis for disease prediction. As a result, feature selection is essential for removing unimportant features. The feature selection significantly minimizes system learning time, which improves CDSS performance efficacy. The MVEFS selects the associated heart disease and diabetes-related features from the clinical dataset. The classifier execution time, accuracy, sensitivity, precision, specificity, and F1-score are the performance metrics used to evaluate the proposed CDSS. According to our experimental study, the MVEFS with a customized deep neural network is more appropriate for predicting heart and diabetes than machine learning algorithms.