Projects
From equations to
explainable models.
Three projects show how I move between mathematical models, controlled computational experiments, and careful interpretation of health-related classifiers.
Each case study separates what has been implemented from what is planned, and gives limitations the same visibility as results.
Selected work
Case studies, not software badges.
The emphasis is on the problem, evaluation design, interpretation, and status—not a wall of tools or a single headline score.
01
Ongoing · Phase 1 implemented · Synthetic-data experiment
Disease-Informed Neural Networks for Epidemic Mobility Models
A PyTorch experiment that fits a SIRD trajectory while enforcing initial conditions and differential-equation residuals. Two-region movement and multipatch SIS models remain planned extensions.
02
Ongoing exploratory · Classification and explainability phase complete · Survival analysis planned
Toward Precision Medicine: Explaining Mortality Risk Predictions in Heart Failure Patients
An exploratory analysis of classifier behavior on a 299-record public dataset, using SHAP to connect global patterns with individual explanations and error analysis.
03
DA 622 Course Project · Completed January–May 2026
Predicting Heart Disease Severity
A comparison of logistic regression, random forest, and weighted XGBoost that makes the severe-class recall–precision tradeoff explicit.
How I read a model
What information did it see?
Data provenance, feature choices, and leakage risk come before performance claims.
Which errors matter?
Evaluation depends on the problem. Accuracy alone can hide the cases that deserve the closest look.
What does the explanation explain?
Attribution methods describe a fitted model’s behavior. They do not establish causality or clinical validity.