Table 2 Linear discriminant analysis (LDA) and random forest (RF) 10-fold cross-validation classification error rates for predicting vocal stimuli emotional category membership.

Stimulus types	Error rates
Stimulus types	LDA	RF	Δ (%)^a
Anna	0.3903	0.3778	3.20
Pseudo-words	0.4597	0.4225	8.09
Semantic positive nouns	0.4112	0.3428	16.63
Semantic negative nouns	0.4614	0.4019	12.90
Semantic neutral nouns	0.4402	0.4438	−0.82
Group Words [across all stimuli (N = 568)]	0.4257	0.3754	11.82
Affect bursts	0.4347	0.4701	−8.14
Pseudo-sentences	0.5211	0.4259	18.27
Lexical sentences	0.5866	0.4448	24.17
Neutral sentences	0.4777	0.4255	10.93
Group Sentences [across all stimuli (N = 470)]	0.5981	0.4813	19.53
Overall [across all stimuli types (N = 1038)]	0.5802	0.4461	23.11

^aThe relative difference between RF error rates and LDA error rates were calculated as follows: (1−(RF error rates/LDA error rates))*100. As it can be observed, the error rates were smaller by RF than LDA, except for semantic neutral nouns and affect bursts. The accuracy rates for both classification methods can be obtained as follows: (1–error rates)*100.

Quick links

Search