Problem

How can health and demographic indicators predict disability at the county level? Can the same model work for both Mobility and Cognitive Disability?

Approach

Cleaned CDC PLACES data covering 2,293 U.S. counties. Trained 1,000+ models in MATLAB Classification Learner using 5-fold cross-validation and a held-out test set. Selected features with mRMR across three tiers.

Result

98.83% test accuracy for Mobility (Cubic SVM), 97.08% for Cognitive (Wide Neural Network). Caught a labeling bug in upstream data that improved accuracy from ~88% to ~99%.