Table 2.
Model performance for the MKL model distinguishing between the different classes of children
| Model MKL | Balanced accuracy (%) | True positives/Total positives | True negatives/Total negatives | AUCROC |
|---|---|---|---|---|
| [TYP] vs. [DD] | 44.05 (P = 0.75) | 18/42 | 19/42 | 0.41 |
| [TYP] vs. [DCD] | 57.50 (P = 0.27) | 14/20 | 9/20 | 0.60 |
| [TYP] vs. [COM] | 75.86 (P = 0.005) | 23/29 | 21/29 | 0.80 |
| [DD] vs. [DCD] | 45.50 (P = 0.76) | 9/20 | 8/20 | 0.51 |
| [DD] vs. [COM] | 46.55 (P = 0.67) | 13/29 | 14/29 | 0.52 |
| [DCD] vs. [COM] | 47.50 (P = 0.61) | 11/20 | 8/20 | 0.41 |
| [TYP] vs. [DCD-COM] | 71.43 (P = 0.001) | 30/42 | 30/42 | 0.75 |
| [TYP] vs. [DD-COM] | 63.10 (P = 0.04) | 28/42 | 25/42 | 0.67 |
All significant MKL models survived FDR correction. Binary classifier performance is summarized through 3 measures: balanced accuracy, true positives/negatives that represent the number of children classified correctly as belonging to class1/class2, and the AUC of the receiver operator characteristic curve (1 represents perfect performance, 0.5 represents random performance).