Table 2.

Comparison of model performance on holdout set

#	Model	Accuracy (%)	F-Score	Sensitivity	Specificity	UAR
1	ADOS-2 total score	94	0.959	0.987	0.833	0.910
2	Highest ranked single feature: a2	89	0.926	0.981	0.645	0.813
3	10 Highest ranked (χ2)	86	0.902	0.902	0.729	0.815
4	10 Highest ranked (Information Gain)	85	0.901	0.920	0.661	0.791
5	10 Highest ranked (Pearson Correlation)	84	0.888	0.871	0.746	0.808
6	Top performing aggregated featurea	87	0.914	0.945	0.661	0.803
7	Top performing cluster + a3, d1, d2, and d5	87	0.914	0.901	0.780	0.844

UAR Unweighted Average Recall (Bone et al. 2015)

^aTop performing aggregated feature: distance between the individual and the non-ASD centroid, which was composed of the means of ADOS-2 codes a2, b1, and b5 for the non-ASD sample