Supervised learning forms the core of many widely used clinical decision-support tools, enabling predictions such as mortality risk, diagnostic assistance, readmission likelihood, and adverse event detection. In this module, you will understand how to convert clinical problems into prediction tasks, define features and labels appropriately, and evaluate whether supervised learning is the right framework for a given healthcare question. The module introduces essential algorithms, including logistic regression, tree-based models, and regularized regression, with a focus on interpretability and clinical reasoning. You will also explore common data pitfalls such as class imbalance and label leakage, both of which can disrupt clinical validity if mishandled. Through practical exercises, you will build foundational models used throughout healthcare analytics.

Unsupervised learning enables clinicians and researchers to uncover hidden structure in patient populations, identify disease subtypes, and discover new risk categories when labeled outcomes are not available. This module focuses on clustering and dimensionality reduction for patient phenotyping, using both structured clinical data and aggregated EHR features. You will explore when and why unsupervised learning is used, compare major clustering algorithms, and practice interpreting clusters. You will also learn dimensionality reduction techniques used to visualize high-dimensional patient data and guide phenotype refinement. Finally, the module covers cluster validation, reproducibility, and clinical interpretability, all of which are essential to safely using unsupervised insights in healthcare.

Healthcare data is inherently temporal, encompassing vitals, lab results, medications, and clinical events collected over time. This module introduces classical and feature-based methods to represent and analyze these longitudinal patterns for early warning, deterioration detection, and forecasting tasks. You will study the challenges of irregular clinical time series, construct time-window-based and aggregation-based features, and apply non-neural sequence modeling techniques suitable for clinical environments. The second half of the module covers rigorous evaluation methods for healthcare models. You will explore discrimination, calibration, thresholding, and clinical utility metrics, and will design validation strategies that respect temporal ordering, avoid information leakage, and reflect real clinical deployment constraints.

In this final module, you will consolidate your learning of supervised learning, unsupervised learning, temporal modeling, and evaluation by completing a hands-on final project. You will complete an end-to-end project involving clinical problem formulation, model development, exploratory analysis, temporal feature construction, and model evaluation. You will justify model choices, articulate assumptions, and interpret findings from a clinical perspective. Emphasis is placed on communication and documentation, ensuring that results can be reviewed by both technical and clinical decision-makers. The module concludes with a course summary, a glossary of key terms, and a final exam designed to assess their conceptual understanding across all modules.