Table 5.

Recognition rates of each activity for different models under subject-dependent validation criteria. ET: ensemble of randomized trees, FC: fully connected layer, AE: autoencoders, and DBN: deep belief networks.

Reference	Input Signals	Segment Length (s)	Feature Extraction	Classifier	Accuracy (%)
Baldominos et al. [45]	Raw signals	60	Handcrafted	ET	95.3
Baldominos et al. [45]	Raw signals	60	CNN hidden layers	FC layer	85
Alo et al. [46]	Raw signals, magnitude vector, pitch and roll vectors	2	Sparse AE layers	FC layer	97.13
Alo et al. [46]	Raw signals, magnitude vector, pitch and roll vectors	2	DBN hidden layers	DBN output layer	91.57
Current model	Raw signals	2.56	Handcrafted	RF	98.7