Diabetes Risk, From Scratch
Four classifiers in NumPy — no scikit-learn
Read the sourceMostly Jupyter Notebook
Results — hover any figure for the baseline and the caveat
4
Models, zero ML libraries
90%+ accuracy on the CDC cohort. The point of this project is the derivation, not the score — scikit-learn would beat it in three lines, which is exactly why writing it by hand was worth doing once.
Overview
A diabetes risk predictor on the CDC Health Indicators dataset where every model is implemented by hand: decision tree, naive Bayes, KNN, and a feed-forward network, all in NumPy. Written this way on purpose — the point was to own the gradient maths and the splitting criteria rather than call .fit() on them.