Trace Brown
I build the model from scratch, then check whether it is telling the truth.
Statistics and biological sciences at NC State, graduating May 2027. Raleigh, NC.
Work
-
Heart failure mortality: from-scratch logistic regression and a bootstrap study
The textbook interval was too narrow for one coefficient. The bootstrap said so, and you can re-run it here.
- Estimate
- Wald 95%
- Percentile 95%
Serum creatinine coefficient from the heart-failure project: 400 bootstrap refits, computed in your browser as the page loads, with bars counting refits per bin. The dashed percentile interval comes out wider than the dotted Wald interval, and that gap is the finding. Re-run it with controls (the report used 1,000 refits). -
Fraud detection on 284,807 card transactions
SQL velocity features and a time-based split, then a classifier scored on what a review team can actually handle.
-
PopperWick: a research system where the models check each other
One model conducts, three from other companies read, cross-examine and verify, and the original document gets the last word. Its backtest bench graded a public demo trade log, charts included.
About
I am a senior at NC State studying statistics and biological sciences, graduating May 2027. The projects here are the ones where I wrote the estimator myself instead of calling a package, so I can say why each number is what it is. The checking is the part I care about most: a confidence interval that is too narrow looks exactly like one that is right, until you test it.
Contact
The fastest way to reach me is email.