Ibrahim Abunadi¹ and Lakshmana Kumar Ramasamy², ¹King Saud University, Saudi Arabia, ²Higher Colleges of Technology, United Arab Emirates

Heart disease is a leading cause of death globally, claiming approximately 17 million lives each year. Often, these deaths are due to heart failure, a condition where the heart cannot supply enough blood to meet the body’s needs. To improve diagnosis and treatment, healthcare professionals increasingly rely on electronic medical records. These records are invaluable for detecting subtle patterns in symptoms and test results that might otherwise go unnoticed. In the realm of medical data analysis, data mining techniques have shown promise in predicting the outcomes of cardiovascular diseases. However, major challenges can occur—overfitting and managing large dimensions of data—can hinder their effectiveness. To address these issues, this paper proposes a novel method that simplifies the data through feature selection, making this model not only more efficient but also easier to understand. Specifically, we introduce a new framework that combines advanced feature selection algorithms (sequential forward and backward, or CSFB) with a blend of traditional machine learning and cutting-edge deep learning techniques. Utilizing algorithms like Support Vector Machine (SVM), K-Nearest Neighbor (KNN), Random Forest (RF), and a deep learning classifier (Dl4jMlpClassifier), this method refines the data to improve predictions of heart disease outcomes. This work findings confirm that this integrated approach -CSFB feature selection combined with the CMD (Combined Machine and Deep learning) algorithm effectively identifies crucial data features and reliably predicts patient survival rates. This advancement holds significant potential for enhancing heart disease diagnostics and patient care strategies.

Heart disease prediction, Cardiovascular disease (CVD) prognosis, Deep learning, Feature selection, Overfitting prevention