Table 2.

SVM classifier and cross validation.

Support vector machine (SVM) classifier
	ASD	Control	Total
Predicted ASD	256	43	299
Predicted control	47	260	307
Total	303	303	606
Accuracy: 85%; Sensitivity: 85%; Specificity: 86%

Four fold cross validation of SVM classifier
	ASD	Control	Total
Predicted ASD	191	108	299
Predicted control	112	195	307
Total	303	303	606
Accuracy: 64%; Sensitivity: 63%; Specificity:64%

A support vector machine classifier using individual age, sex and bilateral habenula volume as input is able to distinguish between ASD and TDC control subjects with 85% accuracy using a balanced dataset created by adding ASD subjects randomly picked from the original 220. The accuracy drops to 64% in a fourfold cross validation where for every quarter of the original dataset, the SVM is trained on the remaining three quarters and applied to the unseen data.

ASD, autism spectrum disorder; TDC, typically developing controls; SVM support vector machine.