Out-of-sample crash prediction on City Crash and Exposure Data (Los Angeles held-out)
-7.43Predictive log-likelihoodGP learned
Evaluation results
| Method | Links | |
|---|---|---|
| GP learnedRandom-effects prior=Gaussian Process, Kernel type=learned embedding2026.06 | -7.43 | |
| Indep RERandom-effects prior=independent hierarchical2026.06 | -7.62 | |
| GP EuclideanRandom-effects prior=Gaussian Process, Kernel type=Euclidean covariate distance2026.06 | -7.99 | |
| GP HDV-onlyRandom-effects prior=Gaussian Process, Kernel type=log-HDV-frequency similarity2026.06 | -8.27 | |
| Single poolRandom-effects prior=none2026.06 | -13.36 |