Skip to main content
. 2021 Jan 4;21(1):291. doi: 10.3390/s21010291

Table 6.

Comparison of the performance of the best systems reproduced from previous works and the best CNN multitask systems.

System Classification Task LOSO AUC Stratified 10-Fold AUC
HAR features [77] + Adaboost (with 100 estimators) Context 0.962 (0.029) 0.970 (0.003)
multitask CNN with raw signal (ours) Context 0.981 (0.017) 3 0.990 (0.002) 3
Hssayeni et al. [15] (Specific tremor features) Tremor 0.922 (0.035) 0.956 (0.003)
multitask CNN with FFT (ours) Tremor 0.936 (0.024) 4 0.965 (0.003) 4

3 Value obtained only from the context classifier output. 4 Value obtained only from the tremor detector output.