Table 1.

Performance metrics for 5 action movements from the UTD-MHAD Dataset.

	Action 4 - Two Hand Front Clap	Action 5 – Right Arm Throw	Action 9 – Draw Circle Clockwise	Action 11 – Draw Triangle	Action 17 – Tennis Serve
RMS	5.616 (0.139)	6.508 (0.172)	6.580 (0.144)	6.098 (0.230)	8.195 (0.208)
Log-likelihood: Autoenconder+Mixture Density Network (GMM)	1.707 (0.569)	0.968 (0.834)	0.488 (0.897)	0.788 (0.467)	0.823 (0.626)
Log-likelihood: Autoenconder+Expectation Maximization (GMM)	1.808 (3.526) 8 states	0.557 (3.481) 8 states	0.429 (5.189) 10 states	0.199 (10.567) 7 states	0.575 (2.999) 7 states
Log-likelihood: Autoenconder+Hidden Markov Model	−0.954 (0.066) 23 states	−0.731 (0.041) 28 states	−0.905 (0.017) 26 states	−1.459 (3.025) 30 states	−0.730 (0.015) 25 states
KL Divergence: Autoenconder+Mixture Density Network (GMM)	Train vs. test data: 1.232	Train vs. test data: 0.685	Train vs. test data: 2.617	Train vs. test data: 1.626	Train vs. test data: 0.709
Heuristic Consistency	0.1002	0.0972	0.0728	0.1053	0.0724
Prediction Intervals	0.0759	0.0966	0.0788	0.0837	0.0559