The advent of machine learning has introduced a paradigm shift in clinical decision support systems. This paper conducts a thorough comparative review of ML algorithms employed for medical diagnosis prediction, encompassing both traditional methods (Naïve Bayes, k-nearest neighbors, support vector machines) and sophisticated deep learning approaches (convolutional neural networks, graph neural networks). Our evaluation, grounded in an analysis of existing literature and performance on benchmarking datasets, focuses on key metrics of accuracy, robustness, and explainability. We further explore predominant methodological trends and significant hurdles, including managing imbalanced data and ensuring model generalizability. Finally, the review outlines a pathway toward creating transparent, dependable, and ethically sound AI solutions for clinical environments.