Comparison of feature importance measures as explanations for classification models

Table 1 Logistic regression feature coefficients and random forest feature importances for the breast cancer data

Feature	Coefficient	Importance	Feature	Coefficient	Importance
Radius 1	− 0.088	0.034	Texture 1	0.103	0.016
Perimeter 1	− 0.061	0.041	Area 1	− 0.020	0.041
Smoothness 1	0.061	0.006	Compactness 1	− 0.702	0.011
Concavity 1	0.831	0.042	Concave points 1	1.114	0.105
Symmetry 1	− 0.062	0.004	Fractal dimension 1	0.031	0.004
Radius 2	2.207	0.014	Texture 2	− 0.396	0.004
Perimeter 2	0.042	0.013	Area 2	0.729	0.032
Smoothness 2	0.211	0.004	Compactness 2	− 0.464v	0.005
Concavity 2	− 0.133	0.007	Concave points 2	0.322	0.005
Symmetry 2	− 0.204	0.004	Fractal dimension 2	− 0.827	0.005
Radius 3	3.170	0.113	Texture 3	1.723	0.019
Perimeter 3	0.482	0.145	Area 3	0.747	0.120
Smoothness 3	0.573	0.013	Compactness 3	− 0.089	0.015
Concavity 3	0.766	0.032	Concave points 3	1.185	0.130
Symmetry 3	0.692	0.010	Fractal dimension 3	0.423	0.006

Bolded are the nine features detected with random forest and nine most important features with regression, ranked based on the p-value