Skip to main content
. 2022 Jun 16;1(6):e0000044. doi: 10.1371/journal.pdig.0000044

Fig 6. Benchmarking Seq2Seq performance.

Fig 6

Prediction outcomes were aggregated from primitives, activities, and patients. Shown are the sensitivity, false discovery rate (FDR), and F1 score for Seq2Seq, convolutional neural network (CNN), action segment refinement framework (ASRF), and random forest (RF). Seq2Seq had the highest sensitivity (Seq2Seq, 0.767; CNN, 0.727; ASRF, 0.720; RF, 0.497). Seq2Seq FDR was lower than CNN and RF but marginally higher than ASRF (Seq2Seq, 0.166; CNN, 0.196; ASRF, 0.160; RF, 0.213). Seq2Seq had the highest F1 score (Seq2Seq, 0.799; CNN, 0.763; ASRF, 0.775; RF, 0.609).