Fig. 5: Prediction interval coverage comparison of conformal deep learning and ECTemp™ models at the subject-session level on the test set. | Communications Engineering

Fig. 5: Prediction interval coverage comparison of conformal deep learning and ECTemp™ models at the subject-session level on the test set.

From: Degrees of uncertainty: conformal deep learning for non-invasive core body temperature prediction in extreme environments

Fig. 5: Prediction interval coverage comparison of conformal deep learning and ECTemp™ models at the subject-session level on the test set.

The left panels show true vs. desired coverage, with the conformal model closely aligning with the perfect coverage line, indicating more reliable uncertainty estimates. The right panels illustrate domain-level coverage, where the conformal model remains consistent across datasets, while ECTemp™ shows greater variability and underestimates uncertainty. Shaded areas reflect subject-session variability. Domain key: Wildland firefighters (WFF, red), race-car drivers (RACE, orange), nuclear plant workers (NUC, green), factory workers (FACT, yellow), mine workers (MINE, black), and explosive ordnance disposal technicians (EOD, purple).

Back to article page