A Comparison Analysis of Heart Disease Prediction Using Supervised Machine Learning Techniques

• Main Authors: Elhadjamor, Emna Ammar, Harbaoui, Houda
• Format: Conference Proceeding
• Language: English
• Subjects: Accuracy; Clinical Decision-Making Support; Diseases; Heart; Heart Disease Prediction; Kaggle Dataset; Machine learning; Machine learning algorithms; Nearest neighbor methods; Prediction algorithms; Random forests; Stochastic processes; Supervised Learning Algorithms; Surveys
• ISSN: 2642-7389
• DOI: 10.1109/ISCC61673.2024.10733656

Evaluating multiple machine learning models for predicting and detecting heart disease is crucial yet challenging within clinical practice. In regions with limited cardiovascular expertise, misdiagnoses are frequent, highlighting the need for precise early-stage prediction using comprehensive analysis of digital patient records. This study aimed to pinpoint the most accurate machine learning classifiers for this pivotal purpose, leveraging a heart disease dataset sourced from the official 2022 annual CDC survey. Thirteen supervised machine learning algorithms underwent rigorous deployment and evaluation to gauge their effectiveness in predicting heart disease. Comparative assessments scrutinized the performance and accuracy of these algorithms, along with estimating the significance of each feature in predicting heart disease. Exploration extended to various ensemble methods and individual classifiers, including AdaBoost, Random Forest, Extra Trees, HistGradientBoosting, Decision Tree, K-Nearest Neighbors (KNN), Multi-layer Perceptron (MLP), Stochastic Gradient Descent (SGD), Logistic Regression, Gaussian Naive Bayes, among others. Particularly noteworthy was the exceptional performance of HistGradient-Boosting, achieving an outstanding in all evaluation metrics. This outcome underscores the potential of a relatively straightforward supervised machine learning approach, hinting at its promising role in enhancing early-stage prediction and detection of heart disease.