Optimized Machine Learning Models to Enhance Aortic Stenosis Severity Classification

This study suggests an optimized machine learning model to enhance the diagnosis of aortic stenosis (AS), a critical cardiovascular disease involving the narrowing of the aortic valve. Timely and accurate diagnosis of AS is needed to decrease morbidity and mortality rates, particularly in the elderly. Conventional diagnostic techniques like echocardiography are accurate but are suffering from inter-observer variation and are expensive. Relying on the strength of machine learning (ML), this paper explores the application of three widely used classification models’ Random Forest (RF), Support Vector Machine (SVM), and k-nearest Neighbors (KNN) to develop a robust, self-sustaining diagnostic tool. We used Bayesian optimization to optimize the hyperparameters of each model on a dense grid of 39 features in 60 patients. These are demographic information, clinical history, and spectral features of heart sound recordings. The data were preprocessed and balanced using SMOTE to facilitate stable training of the models. The findings reveal that the optimal KNN model performed better compared to others regarding accuracy, sensitivity, specificity, and F1 score. These results underscore the promise of optimal ML models to aid clinicians with precise and efficient AS diagnosis and create opportunities for future application in clinical practice.