Machine Learning for Healthcare

MIT 6.S897/HST.956 Machine Learning for Healthcare (Spring 2019), co-taught by Professors David Sontag and Peter Szolovits, is one of the most authoritative free graduate courses on applying ML to real clinical problems, and the full 25-lecture video set plus lecture notes, readings, and project guidelines are freely available on MIT OpenCourseWare under a Creative Commons (CC BY-NC-SA) license. Rather than re-teaching ML fundamentals, it focuses on what makes healthcare data hard: messy EHR and insurance-claims data, risk stratification, survival modeling, clinical NLP, causal inference, disease progression and subtyping, fairness, dataset shift, and regulation, anchored by guest lectures from practicing Boston-area clinicians. It assumes a prior ML course as a prerequisite, so it is genuinely advanced and not an entry point. As a self-study resource it has real limitations: it is a 2019 archive (predating the modern LLM/foundation-model wave), there is no certificate, no graded feedback or autograder, and the original problem sets relied on restricted datasets (MIMIC, IBM MarketScan) that require separate credentialed access. For a machine-learning practitioner or clinician-researcher who already knows the math and wants domain depth, it remains an excellent, citable free resource; for beginners it is the wrong starting point.

Outstanding, free, expert-taught domain course on clinical ML, but only worthwhile if you already have ML fundamentals; it is a 2019 archive with no certificate, no graded feedback, and problem sets that depend on access-restricted clinical datasets.

Best for: ML engineers, data scientists, and graduate students who already understand supervised learning, probability, and (ideally) some deep learning, and who want a rigorous, application-driven understanding of how ML is actually used on clinical data. It is especially valuable for people moving into health-tech or clinical-AI research, and for clinicians or biomedical researchers with a strong quantitative/ML background who want to learn the modeling side and the failure modes (bias, confounding, dataset shift) specific to healthcare.

Skip if: Complete beginners or anyone without a prior machine learning course. MIT explicitly lists a prior ML class as a prerequisite (e.g., 6.036/6.862, 6.867, 9.520, 6.806/6.864, 6.438, or 6.034), and lectures assume that background. It is also a poor fit for people who want a certificate, hands-on graded labs with feedback, or up-to-date coverage of LLMs and foundation models in medicine, since this is a static Spring 2019 recording.