Deep-learning: investigating deep neural networks hyper-parameters and comparison of performance to shallow methods for modeling bioactivity data

Table 4 Performance achieved by DNN, NB, kNN, RF and SVM measured using MCC as evaluation metric

Dataset	Algorithm	MCC 1st	MCC 2nd	MCC 3rd	Mean MCC	Std MCC
DRD4	DNN	0.900	0.895	0.867	0.887	0.018
	SVM_rbf	0.889	0.876	0.865	0.876	0.012
	SVM_linear	0.828	0.816	0.840	0.828	0.012
	RF	0.867	0.854	0.862	0.861	0.007
	kNN	0.762	0.778	0.763	0.767	0.009
	NB	0.742	0.761	0.750	0.751	0.009
HERG	DNN	0.858	0.913	0.880	0.884	0.028
	SVM_rbf	0.845	0.891	0.872	0.869	0.023
	SVM_linear	0.773	0.782	0.780	0.778	0.005
	RF	0.838	0.857	0.848	0.847	0.010
	kNN	0.818	0.813	0.825	0.819	0.006
	NB	0.612	0.620	0.602	0.611	0.009
CDK2	DNN	0.919	0.932	0.927	0.926	0.007
	SVM_rbf	0.920	0.913	0.922	0.919	0.005
	SVM_linear	0.863	0.889	0.864	0.872	0.015
	RF	0.895	0.912	0.902	0.903	0.008
	kNN	0.895	0.904	0.910	0.903	0.007
	NB	0.769	0.773	0.780	0.774	0.006
CogX	DNN	0.982	0.981	0.987	0.983	0.003
	SVM_rbf	0.978	0.979	0.980	0.979	0.001
	SVM_linear	0.971	0.970	0.977	0.973	0.004
	RF	0.973	0.979	0.982	0.978	0.004
	kNN	0.968	0.971	0.971	0.970	0.002
	NB	0.889	0.897	0.882	0.889	0.008
CYP_19A1	DNN	0.893	0.920	0.899	0.904	0.014
	SVM_rbf	0.886	0.896	0.889	0.890	0.005
	SVM_linear	0.849	0.866	0.862	0.859	0.009
	RF	0.873	0.910	0.879	0.887	0.020
	kNN	0.805	0.811	0.821	0.812	0.008
	NB	0.755	0.821	0.775	0.784	0.034
CB1	DNN	0.943	0.941	0.940	0.941	0.002
	SVM_rbf	0.941	0.937	0.931	0.936	0.005
	SVM_linear	0.885	0.893	0.881	0.886	0.007
	RF	0.908	0.923	0.914	0.915	0.008
	kNN	0.906	0.921	0.901	0.909	0.011
	NB	0.758	0.781	0.765	0.768	0.012
CAII	DNN	0.858	0.885	0.843	0.862	0.021
	SVM_rbf	0.857	0.851	0.866	0.858	0.007
	SVM_linear	0.828	0.826	0.830	0.828	0.002
	RF	0.836	0.857	0.861	0.851	0.013
	kNN	0.558	0.557	0.577	0.564	0.011
	NB	0.754	0.769	0.783	0.769	0.015

Results for each activity class and validation set from three experiments as shown. Best recorded results for each activity class are highlighted in italic

ISSN: 1758-2946