Machine Learning in Cardiology: Assessing Model Reliability for Real world Deployment

Abstract

Cardiovascular disease remains one of the leading causes of death worldwide and there is an imperative need to have strong early-identification strategies. This paper makes a rigorous comparison of the five commonly used machine learning algorithms- Logistic Regression, Decision Tree, Random Forest, Support Vector Machine and K-Nearest Neighbors used to analyze the UCI Heart Disease dataset. The preprocessing of the dataset involved the traditional method of scaling and the separation of the data into training and testing samples at 70:30 proportion. They were then used to determine predictive efficacy in terms of accuracy, precision, recall, and F1-score. Random Forest performed the best in predictive accuracy, recording 90.16 %. Logistic Regression and SVM were also really close to each other in terms of performance with accuracies of 85.25 % and 81.97% respectively. The figures obtained in Decision Tree and KNN were 78.68 % and 75.41 in terms of accuracy respectively. The results, therefore, demonstrate that ensemble-based classifiers, especially Random forest, have better performances in clinical settings, and support the prevailing effect selection of algorithm in healthcare analytics. Due to its shown robustness and ability to support complex data interactions, Random Forest becomes a potentially highly effective model to use in a real-world scenario. Later research could usefully look into hybrid models or deep-learning techniques that would succeed further in terms of diagnostic reliability.

Ähnliche Arbeiten