Semisupervised Deep Features of Time-Frequency Maps for Multimodal Emotion Recognition

Table 9

The effect of feature reduction on the accuracy.

Method	Scenario
	Four-class		Two-class
	Acc	Kappa	Acc	Kappa
Without feature reduction	0.806	0.612	0.831	0.662
PCA	0.837	0.674	0.859	0.718
LDA	0.852	0.704	0.872	0.744
PCA + LDA	0.873	0.746	0.897	0.794
Static SSDR	0.869	0.738	0.885	0.770
Optimized SSDR	0.928	0.856	0.953	0.906

The bold values represent the highest accuracies.