Table 5.

Performance of the committee of 15 greedy classifiers trained using all controlling fixations as target fixations (Trainset 2).

Sbj.	Test on fixations with 500 ms threshold					Test on fixations with 1000 ms threshold
	Specificity	Sensitivity				Sensitivity
		All	Button	Ball	Free cell	Button	Ball	Free cell
1	0.92 ± 0.08	0.22 ± 0.02	0.28 ± 0.07	0.22 ± 0.05	0.15 ± 0.03	0.25	0.27	0.17
2	0.91 ± 0.05	0.24 ± 0.08	0.30 ± 0.06	0.23 ± 0.09	0.18 ± 0.12	0.21	0.17	0.14
3	0.96 ± 0.04	0.18 ± 0.03	0.18 ± 0.04	0.19 ± 0.05	0.18 ± 0.09	0.11	0.10	0.12
4	0.91 ± 0.04	0.20 ± 0.02	0.30 ± 0.06	0.19 ± 0.01	0.10 ± 0.05	0.21	0.20	0.09
5	0.93 ± 0.06	0.26 ± 0.06	0.23 ± 0.11	0.24 ± 0.06	0.30 ± 0.09	0.22	0.24	0.19
6	0.93 ± 0.03	0.33 ± 0.06	0.34 ± 0.12	0.29 ± 0.04	0.36 ± 0.09	0.24	0.19	0.32
7	0.92 ± 0.08	0.26 ± 0.04	0.31 ± 0.11	0.27 ± 0.02	0.19 ± 0.06	0.29	0.21	0.17
8	0.84 ± 0.06	0.45 ± 0.07	0.50 ± 0.10	0.42 ± 0.12	0.43 ± 0.09	0.46	0.39	0.37
Mean	0.91	0.27	0.31	0.26	0.24	0.25	0.22	0.20
Std.	0.07	0.10	0.09	0.07	0.13	0.10	0.08	0.10

For training, recordings with 500 ms dwell threshold were used. On these recordings, 5-fold cross-validation was used to estimate, as M ± SD over folds, all the performance indices. For each fold, the classifier was trained on 80% and tested on 20% of the set. In the case of testing on 1000 ms dwell threshold data, the train and test sets were not overlapped and, therefore, the classifier was applied to all test data.