Table 2.

Performance of the proposed bi-modular sequential neural network with different optimization methods.

Optimizer Selection Experiment
Optimizer	Class AP (angry)	Class AP (happy)	Class AP (sad)	Class AP (neutral)	Micro-mAP
Adam	0.991	0.802	0.694	0.360	0.915
RMSprop	0.995	0.904	0.835	0.460	0.955
SGD	0.876	0.436	0.217	0.265	0.679

RMSprop achieves the best mean and class average precision scores.