HEALTH DESK · ★ HONORABLE MENTION
MODEL SURFACES CARE DISPARITIES ACROSS 75 SUBGROUPS
Three complementary models expose who is being left behind by healthcare access — and where it is heading
SEATTLE — Healthcare access barriers fall unevenly across demographic subgroups, in ways invisible to aggregate statistics. Working from a single dataset spanning 2019 through 2025, this datathon entry built not one model but three — predictive, time-series, and clustering — each answering a different question: what is predictable, where trends are heading, and who is being missed. The analysis earned an honorable mention.
The predictive arm trained six supervised algorithms — linear, ridge and lasso regression, decision trees, random forest, and gradient boosting — on a 70/30 split to score cost barriers by subgroup, reaching 93.7 percent accuracy with an average error of roughly half a percentage point.
HOW IT WORKS
A forecasting pipeline of ARIMA-style models, exponential smoothing, moving averages and polynomial regression projected a 9.15 percent average barrier rate for 2025, within a 95 percent confidence interval of 7.93 to 10.36. Mental health emerged as the only category still worsening.
The third arm applied K-Means, hierarchical clustering, DBSCAN, Isolation Forest and Local Outlier Factor with PCA reduction, flagging 11 high-confidence anomalies and 19 at-risk subgroups — populations such as Native Hawaiian and Pacific Islander individuals, statistically buried in aggregates yet facing barriers of 14 to 23 percent. All three models run end to end in about 90 seconds.
LESSONS FROM THE FLOOR
The data held a paradox: barriers decreased in 2020, likely on expanded coverage and telehealth, then climbed sharply after 2022 as policies rolled back. A single aggregate metric would have missed both that reversal and the mental-health divergence — the clearest argument, the builder notes, for letting three models tell one story.
Source available → github.com/Aditya9246 ↗
CONTINUED: Zero-shot vision verifies beach cleanups, Page A4 →