Table 5. Unsupervised CEERM recognition performance based on different numbers of RBM layers.

Model+Abstract Layers	R	P	F	FI	FIR(%)
DBN1+POSL+DPL+LENL+LOCL+DISL+TWFL	80.4	78.5	79.44	0	0
DBN2+POSL+DPL+LENL+LOCL+DISL+TWFL	83.2	78.7	80.89	1.45	1.82
DBN3+POSL+DPL+LENL+LOCL+DISL+TWFL	84.7	82.8	83.74	2.85	3.53
DBN4+POSL+DPL+LENL+LOCL+DISL+TWFL	87.32	83.12	85.17	1.43	1.71
DBN5+POSL+DPL+LENL+LOCL+DISL+TWFL	86.13	83.21	84.64	−0.52	−0.61
DBN6+POSL+DPL+LENL+LOCL+DISL+TWFL	85.21	81.21	83.16	−1.48	−1.75
DBN7+POSL+DPL+LENL+LOCL+DISL+TWFL	83.13	81.32	82.22	−0.95	−1.14
DBN8+POSL+DPL+LENL+LOCL+DISL+TWFL	82.59	80.12	81.34	−0.88	−1.07

Note: F_n is the F measure when the number of RBM layers is n, FI is difference between F_n and F_n−1,

FIR is equal to $\frac{F_n-F_{n-1}}{F_{n-1}}$.