An expert dataset of 4803 annotated chest compressions from 10 experts was collected and downsampled to 3434 samples to train the mistake-detection models¶

Claim · All claims
Evidence · 1 study · design r2 · q2
Currency · newest study 2022 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? The expert group dataset contained 4803 chest compressions, each annotated with five target classes, and was downsampled to 3434 samples because of class imbalance. → Di Mitri 2022

Evidence¶

Di Mitri 2022¶

Di Mitri, D., Schneider, J., & Drachsler, H. (2022). KeepMeInTheLoop: Real-Time Feedback with Multimodal Data. International Journal of Artificial Intelligence in Education (2022) 32:1093–1118. https://doi.org/10.1007/s40593-021-00281-z

q2 · i? · design · r2

Expert data collection phase with 10 participants performing four 1-minute sessions; the article reports the dataset "consisted of 4803 CCs" with five annotations per compression, later downsampled to 3434 samples (-28.5%) due to unbalanced classes.

"The collected dataset from the expert group consisted of 4803 CCs. Each CC was annotated with five classes."

Discussion¶

Related Claims¶

• Sessions with CPR Tutor feedback showed no significant difference in overall mistake frequency compared with control sessions without feedback — related
• Dialogue-shuffling augmentation improves evaluator performance while downsampling reduces it — related
• The CPR Tutor architecture classifies a chest compression in about 70 milliseconds, supporting real-time feedback provision — related
• Real-time audio feedback from the CPR Tutor is followed by a short-term decrease in Error Rates across all five target CPR classes within 10 seconds — related