My M.S. Data Science capstone — a predictive modeling pipeline to assess individual health risk factors using real-world clinical datasets.
Identifying patients at risk before health events occur is one of the most high-impact applications of predictive modeling — but it requires careful handling of noisy, complex clinical data.
Built a full ML pipeline: data cleaning, feature engineering from clinical variables, model selection and tuning, and evaluation with healthcare-appropriate metrics (AUC-ROC, precision-recall).
A validated predictive model achieving [your metric] on held-out test data, with an interpretability layer to make predictions explainable to clinical stakeholders.
This was my capstone for the M.S. in Data Science at Boston University — a project where I could bring together everything I'd learned across the program: statistical modeling, machine learning, feature engineering, and how to communicate findings to a non-technical audience.
Healthcare data is among the messiest and most important in data science. Missing values, class imbalance, and the real-world cost of false negatives all had to be accounted for. I spent significant effort on feature engineering from raw clinical variables and ensuring the model's predictions were interpretable — not just accurate.
Add your specific dataset, the health outcome you were predicting, and any domain-specific decisions you made here.
Data preprocessing — Handling missing values, outlier treatment, and normalization of clinical features.
Feature engineering — Derived risk indicators from raw clinical variables; handled class imbalance with [SMOTE / reweighting / etc.].
Model selection — Compared [add your models: logistic regression, random forest, XGBoost, etc.] with cross-validation.
Interpretability — Used [SHAP / feature importance / etc.] to make predictions explainable.