AAI_2025_Capstone_Chronicles_Combined

Models 1 and 2 exhibit similar calibration quality on the in-location test set, both outperforming Model 0. The uncalibrated ECE scores of ~2.5% indicate well-calibrated models according to established benchmarks (Pavlovic, 2025). Calibration improved Model 2's performance while having no effect on Model 1. On the out-location test set, calibration degraded substantially: uncalibrated ECE values reached 10% (Model 1) and 8% (Model 2). Post-calibration ECE improved to ~6.5% for both models, though this indicates moderate calibration quality at best.

294