Table 2.

The performance of MAM in internal cross-validation and external validation

Method	AUC	Accuracy	Sensitivity	Specificity	PPV	NPV
Internal cross-validation
EML	0.846 ± 0.034	0.787 ± 0.017	0.786 ± 0.078	0.788 ± 0.039	0.556 ± 0.033	0.918 ± 0.026
STAM	0.880 ± 0.005	0.840 ± 0.020	0.832 ± 0.030	0.842 ± 0.036	0.643 ± 0.048	0.938 ± 0.008
WO-GMA	0.912 ± 0.010	0.856 ± 0.017	0.879 ± 0.042	0.848 ± 0.025	0.663 ± 0.034	0.955 ± 0.015
MAM	0.973 ± 0.007	0.938 ± 0.007	0.939 ± 0.021	0.934 ± 0.014	0.826 ± 0.031	0.980 ± 0.006
MAM.w/o.info	0.965 ± 0.006	0.931 ± 0.012	0.912 ± 0.010	0.937 ± 0.016	0.832 ± 0.034	0.969 ± 0.003
External validation
EML	0.844 ± 0.026	0.767 ± 0.027	0.850 ± 0.035	0.744 ± 0.039	0.483 ± 0.037	0.947 ± 0.010
STAM	0.882 ± 0.011	0.810 ± 0.016	0.879 ± 0.018	0.791 ± 0.025	0.539 ± 0.025	0.959 ± 0.004
WO-GMA	0.906 ± 0.014	0.848 ± 0.019	0.904 ± 0.048	0.832 ± 0.035	0.603 ± 0.040	0.970 ± 0.013
MAM	0.967 ± 0.005	0.934 ± 0.008	0.925 ± 0.024	0.936 ± 0.009	0.802 ± 0.022	0.978 ± 0.008
MAM.w/o.info	0.966 ± 0.006	0.928 ± 0.008	0.908 ± 0.038	0.933 ± 0.005	0.790 ± 0.013	0.974 ± 0.011

Data are represented by mean ± sd. The highest value for each metric is shown in bold.

EML ensemble machine learning model by Mccay et al.³⁰, STAM spatio-temporal attention-based model by Nguyen-Thai et al.³¹, WO-GMA weakly supervised online action detection model by Luo et al.³², MAM motor assessment model, MAM.w/o.info MAM without the Info Branch, AUC area under receiver operating characteristic curve, PPV positive predictive value, NPV negative predictive value.