Table 2.

Activity classification performance ${}^1$.

Model	MHEALTH	WISDM	SPAR
FCN	0.925 ± 0.049	0.754 ± 0.012	0.947 ± 0.069
PEF	0.984 ± 0.029	0.852 ± 0.060	0.971 ± 0.038
PDF	0.995 ± 0.016	0.889 ± 0.055	0.980 ± 0.028
PTN ${}^{†}$	0.993 ± 0.024	0.909 ± 0.054	0.978 ± 0.035
PTN	0.999 ± 0.003	0.913 ± 0.053	0.990 ± 0.017

¹ Classification performance of the fully-convolutional neural network (FCN), personalized engineered feature model (PEF), personalized deep feature model (PDF), personalized triplet network trained with conventional triplet loss (PTN^†), and the personalized triplet model trained with patient-specific triplet loss (PTN). Scores are the cross-validated classification accuracy (mean ± standard deviation) aggregated by subject. ^† The PTN trained with conventional triplet loss.