Why Comparison Is Hard
- Different datasets and preprocessing.
- Different metrics and evaluation.
- Strong results are not always comparable.
Our response
One dataset, one split, one protocol.
Apples vs Oranges
Paper A
TaskNarrow target
DatasetPrivate cohort
SignalsSingle/few channels
MetricAccuracy
Paper B
TaskBroader detection
DatasetPublic benchmark
SignalsMultichannel PSG
MetricRanking metric
Paper C
TaskScreening classification
DatasetAnother cohort
SignalsEngineered features
MetricSensitivity / specificity
Our response
DatasetSame
SplitsSame
Preproc.Same
1. Common dataset
2. Common splits + labels
3. Whole-pipeline comparison