Using this framework, we ran a comparative study of ECG arrhythmia classification. In this folder we provide more extensive classification scores.
To demonstrate the difference between inter- and intra-patient evaluation, we have provided results (test set F1 score) for form classification on three beat-level datasets in the table summary_form_evaluation_inter_vs_intra.csv.
These results were obtained with cross-validation. From these numbers, we can observe how the intra-patient evaluation schema gives much higher scores. Considering that intra-patient is not a realistic scenario, it demonstrates that it is an overly optimistic evaluation.
To give a more comprehensive overview of arrhythmia classification, we provide classification scores on the train, validation and test sets. This includes per-class metrics (F1 and AUC), as well as average scores over all classes (macro average).
Form classification results on beat-level datasets: summary_form_beatlevel.csv
Form classification results on recording datasets: summary_form_recordinglevel.csv
Rhythm classification results on beat-level datasets: summary_rhythm_beatlevel.csv
Rhythm classification results on recording datasets: summary_rhythm_recordinglevel.csv