Feature Selection and Classification of Coronary Artery Disease using Machine Learning and Deep Learning Techniques

Abstract

Coronary artery disease (CAD), particularly within the broader category of Cardiovascular newlineDiseases (CVD), poses a severe threat to wellness worldwide, affecting millions of people and giving rise to a significant annual fatality count. The increasing newlineprevalence of CAD not only impacts individuals living conditions but also places substantial pressure on medical resources worldwide. Precise and prompt forecasting of CAD is Vital for prompt intervention and medication, potentially cutting down the mortality rate and improving patient outcomes. Despite numerous studies aimed at predicting heart disease, many have failed to account for the complex interplay of various risk factors, often relying on limited datasets that constrain the accuracy and applicability of their results. newlineThis thesis introduces various methodologies to strengthen the prediction and classification of CAD using large and diverse datasets by applying Machine Learning (ML) newlineand Deep Learning (DL). By dealing with the constraints of earlier research, this study newlineleverages extensive datasets to develop robust predictive models. The first study explores the creation of new synthetic features, during forecasting. This research used algorithms such as XGBoost, Naive Bayes, AdaBoost, Logistic Regression, LGBoost, newlineand Stochastic Gradient Descent, are utilized. Through the HyperOpt method, performance enhancement is accomplished via parameter optimization. The model is assessed newlineutilizing the CVD dataset from Kaggle which contains over 70,000 records. newlineThe second study proposes the TLV (Two-Layer-Voting) model, an ensemble method newlinecomparing the efficacy of soft voting and hard voting. In the first layer advanced feature selection techniques, including the ANOVA f-test, Chi-squared test, and Mutual newlineInformation, are combined to improve prediction accuracy. The second layer includes techniques such as Random Forest, Multi-Layer Perceptron, Decision Tree, and Support Vector machine algorithm. The TLV model is tested on both Kaggle s CVD dataset newlineand UCI s heart di