Table 2.

Results of out-of-sample validation.

Set	% pos	Combo	Roc	Avg precision	Accuracy	Precision	Recall
10X	26.90%	distances	0.581	0.295	0.727	0.307	0.230
		Dist-atr	0.566	0.265	0.739	0.315	0.197
		Atchley	0.669	0.443	0.807	0.742	0.132
		Atchley-dist	0.616	0.359	0.782	0.459	0.163
		Atchley-dist-atr	0.592	0.322	0.773	0.406	0.159
Dash	7.33%	distances	0.605	0.108	0.740	0.111	0.362
		Dist-atr	0.650	0.124	0.802	0.139	0.326
		Atchley	0.690	0.183	0.910	0.237	0.104
		Atchley-dist	0.611	0.180	0.799	0.133	0.316
		Atchley-dist-atr	0.648	0.147	0.824	0.154	0.311
Expt	12.70%	distances	0.733	0.332	0.730	0.275	0.688
		Dist-atr	0.707	0.454	0.714	0.250	0.625
		Atchley	0.809	0.698	0.786	0.333	0.688
		Atchley-dist	0.807	0.667	0.738	0.270	0.625
		Atchley-dist-atr	0.768	0.532	0.722	0.256	0.625
Atlas	86.60%	distances	0.463	0.852	0.840	0.866	0.964
		Dist-atr	0.504	0.863	0.777	0.864	0.881
		Atchley	0.570	0.897	0.866	0.866	1.000
		Atchley-dist	0.471	0.867	0.863	0.867	0.993
		Atchley-dist-atr	0.497	0.869	0.834	0.866	0.957
NewVdj	0.70%	distances	0.528	0.010	0.832	0.009	0.205
		Dist-atr	0.535	0.009	0.908	0.013	0.159
		Atchley	0.516	0.008	0.981	0.014	0.023
		Atchley-dist	0.549	0.010	0.953	0.004	0.023
		Atchley-dist-atr	0.559	0.009	0.953	0.000	0.000

Results of predicting the validation sets with the model trained on the STCRDab set, using different subsets of features. In each section, the best-performing model is highlighted in bold and underlined. The precision and recall are measured for the optimum SVC hyperplane giving the highest AUC.