Heart disease is among the top causes of death globally, thus accurate and timely prediction is critical for improving patient outcomes. This article examines both supervised and semi-supervised learning techniques for heart disease prediction, highlighting the significance of developing precise models for early diagnosis and treatment. Supervised models such as KNearest Neighbors (KNN), Logistic Regression, and Support Vector Machines (SVM) were assessed, achieving accuracies of 80%, 89%, and 87 %, respectively. Additionally, semi-supervised methods, including FixMatch and a Generative Adversarial Network (GAN), were evaluated, with FixMatch achieving an accuracy of 84 % and GAN reaching 82 %. The study concludes that supervised learning, especially Logistic Regression, remains a dependable strategy for predicting heart disease. Nonetheless, semi-supervised techniques like FixMatch show promise in scenarios where labeled data is limited, indicating opportunities for further research and optimization of models for clinical use.