Improving Cardiovascular Disease Prognosis Using Outlier Detection and Hyperparameter Optimization of Machine Learning Models

Abstract

Cardiovascular diseases, globally recognized as prominent contributors to morbidity and mortality, have led to an imperative demand for precise, accessible, and efficient diagnostic methodologies. This study introduces a hybrid classification system integrating an ensemble model and a Fuzzy C Means-based neural network with the objective of augmenting predictive accuracy. A comparative analysis on scalar standards was undertaken to determine the optimal feature scaling technique, thereby enhancing predictive proficiency while optimizing time efficiency. The study further incorporates Random Forest, Support Vector Machines, k-Nearest Neighbor, and deep learning models into the diagnostic framework, while employing a confusion matrix as a performance evaluation tool. The GridsearchCV technique is utilized for hyperparameter optimization, its influence on the accuracy of machine learning (ML) models is critically examined. Special attention is given to the role of outliers and their manipulation using supervised ML algorithms, investigating the impact of outlier exclusion on model accuracy. The experimental data was sourced from a cardiovascular patients dataset in the UCI Machine Learning Repository. The findings of the study suggest that the proposed classifier ensemble model surpasses comparable advancements, achieving an exemplary classification accuracy of 98.78%. This paper thus contributes to the evolving landscape of ML application in cardiovascular disease prediction, emphasizing the significance of outlier detection and hyperparameter optimization.