Table 2.
Confusion Matrix and Estimates of Area Under the Curve (AUC) Based on Validations of the Fitted Multiclass Classifier on Two Test Datasets
| Validation on Test-1 | Reference | AUCb | |||||||
|---|---|---|---|---|---|---|---|---|---|
| Prediction | CON | SCZ | BPAD | MDD | Group | CON | SCZ | BPAD | MDD |
| CON | 20 | 4 | 4 | 6 | CON | 0.85 | |||
| SCZ | 7 | 18 | 5 | 3 | SCZ | 0.84 | 0.85 | ||
| BPAD | 4 | 3 | 17 | 6 | BPAD | 0.85 | 0.82 | 0.78 | |
| MDD | 4 | 5 | 9 | 18 | MDD | 0.82 | 0.84 | 0.69 | 0.76 |
| Validation on Test-2 a | Reference | AUC b | |||||||
| Prediction | CON | SCZ | BPAD | MDD | Group | CON | SCZ | ||
| CON | 115 | 17 | CON | 0.64 | |||||
| SCZ | 24 | 35 | SCZ | 0.77 | 0.89 | ||||
| BPAD | 31 | 1 | |||||||
| MDD | 21 | 0 |
Note: CON, Healthy Control; SCZ, Schizophrenia; BPAD, Bipolar Affective Disorder; MDD, Major Depressive Disorder.
aThe validation on Test-2 dataset does not have any representation of BPAD and MDD patients.
bAUC table represents the overall groupwise AUC by one-versus-all (OVA) (diagonal element) and pairwise AUCs by one-versus-one (OVO) methods (off-diagonal elements).
The bolds in the table on the left confusion matrix are the correctly classified cases,on the AUC right side the bold are one versus all comparisons,the non bold are One versus one comparisons.