DefinePK

Abstract

Heart disease popularly known as cardiovascular diseases (CVDs), continue to rank among the world's top causes of death, demanding the great need of designing accurate and effective diagnostic techniques. Early detection and accurate assessment of heart conditions can significantly enhance patient outcomes and lower the cost of healthcare. In this paper we have investigated the impact of machine learning (ML) models that use ECG data to predict heart disorder/diseases are affected by advanced feature selection techniques. Utilizing a dataset of 986 patients, the study focused on important features extracted using methods namely- Mutual Information (MI), Recursive Feature Elimination (RFE), L1 Regularisation, and Principal Component Analysis (PCA). The deep learning models involving are CNN, LSTM, MLP and ViT were investigated. With an accuracy of 99.30%, CNN with MI produced results that were competitive among the evaluated configurations, while LSTM with MI showed the best accuracy of 99.07%. With the ten feature selection approach-Top, CNN becomes the best model there also for the task at hand with the accuracy of 99.07%. The MLP and ViT models also performed well, achieving high precision (~97.74%) and accuracy of 97.67%. This comparative analysis emphasizes the significance of feature selection in reducing dimensionality, increasing computing effectiveness, and improving the model’s performance. These finding highlight the potential of integrating advanced feature selection techniques with machine learning algorithms to detect cardiac disorders early. In order to increase clinical usability and ensure reliable performance across a variety of datasets, future research may explore hybrid models and real-time applications.