Table 2. Classification results with different settings.

	Mutation feature				Mutation feature + Personal features
Number of Hidden Nodes	Training Time for each fold (seconds)	Testing Time for each fold (seconds)	Training Accuracy (average)	Testing Accuracy (average)	Training Time for each fold (seconds)	Testing Time for each fold (seconds)	Training Accuracy (average)	Testing Accuracy (average)
50	0.0144	0.0002	0.8155	0.7917	0.0145	0.0004	0.8004	0.8333
100	0.0379	0.0004	0.8155	0.7262	0.0516	0.0002	0.9303	0.8810
150	0.0555	0.0003	0.8155	0.7024	0.1147	0.0003	0.9762	0.9583
200	0.0767	0.0010	0.8155	0.6845	0.1461	0.0006	0.9762	0.9286
250	0.0898	0.0006	0.8155	0.6607	0.1649	0.0005	0.9762	0.9226
300	0.1010	0.0005	0.8156	0.6607	0.1642	0.0005	0.9763	0.9107
350	0.1076	0.0009	0.8155	0.6607	0.1729	0.0007	0.9762	0.9048
400	0.1142	0.0010	0.8154	0.6488	0.1827	0.0007	0.9763	0.8869
450	0.1229	0.0006	0.8156	0.6488	0.1887	0.0009	0.9762	0.8869
500	0.1323	0.0010	0.8155	0.6488	0.1978	0.0009	0.9763	0.8810

This table shows the prediction results of the response level to the specific inhibitors for the observed patients. Two feature sets, one including the mutation feature only while the other involving both the mutation feature and personal features, are applied for a comparison. Extreme learning machines and leave-one-out cross validation are used in the calculation. The number of hidden nodes varies from 50 to 500 at a step of 50, and the execution time and accuracy are shown in the table.