Table 3.

Performance of MCPD-Net when 50% of the silhouette and accelerometer data are missing. This demonstrates an overall improvement by our proposed method.

	Precision	Recall	F1-Score
(a) Missing Sil (Only Using Acl)
Acl VAE (unimodal)	0.69	0.66	0.58
AE with $L_D$ (multimodal)	0.61	0.40	0.46
VAE without $L_D$ (multimodal)	0.63	0.62	0.57
MCPD-Net	0.70	0.77	0.64
(b) Missing Acl (only Using Sil)
Sil VAE (unimodal)	0.57	0.63	0.59
AE with $L_D$ (multimodal)	0.67	0.42	0.48
VAE without $L_D$ (multimodal)	0.58	0.61	0.55
MCPD-Net	0.63	0.63	0.63