Unmasking the chameleons: A benchmark for out-of-distribution detection in medical tabular data

Mohammad Azizmalayeri, Ameen Abu-Hanna, Giovanni Cinà

Abstract: Highlights•We provide an open source benchmarking framework for out-of-distribution detection methods in medical tabular datasets.•The framework is applied to large public datasets and allows for identifying the top-performing OOD detection methods.•Our results provide new insights into how methods perform in different conditions.•We demonstrate the challenge posed by over-confidence in prediction models for detecting out-of-distribution instances.