Extended Data Fig. 5: SVM decoding accuracy after accounting for odors with high accuracy.
From: The human brain modulates sniffs according to fine-grained perceptual features of odours
The left three panels show histograms of mean accuracy (number of trials with significant prediction/total number of trials) per odor for 3 subjects. The right panel shows decoding accuracy (percentage of trials reported correctly) of a multiclass SVM model that predicts (out of sample) odor identity based on the features of the sniff at each trial, after removing odors that show high prediction accuracy (> 0.2) as seen in the left panels. Decoding was significant for all subjects (mean subject: classification performance = 2.86%, P = 0.000, one tailed z-test). Data are presented as mean values and error bars indicate 95% C.I.
