Classification accuracies for the word generation task.

Columns two through four show accuracies when 1–3 features were selected from the entire candidate pool. Column 5 shows the accuracies when the candidate pool was restricted to the 16 bilateral features. In this case, the analysis was performed for only the selection of three features, using the same feature selection algorithm. The final column shows classification accuracies when only respiratory features were used. All participants except for Participant 8 displayed significantly higher accuracies for three-feature TCD classification (using the entire candidate pool) than for classification based on respiration. Mean classification accuracy was significantly greater for two and three features than for one feature (repeated-measures regression, p0.012). The comparison between two and three features approached significance (p = 0.056).