Table B1. Open in new tab Feature...

Table B1.

Feature importance for RF classification with unweighted training and testing for different numbers of classes corresponding to the different depth of the class separation in the bottom panel of Fig. 2. Depths 1, 2, 3, and 4 correspond to two-, four-, five-, and six-class classifications, respectively.

Feature	Depth 1	Depth 2	Depth 3	Depth 4
LP_index1000MeV	0.159	0.167	0.160	0.163
log10(Signif_Avg)	0.077	0.128	0.140	0.142
log10(Variability_Index)	0.100	0.114	0.114	0.113
log10(Unc_Energy_Flux100)	0.128	0.109	0.105	0.108
LP_beta	0.103	0.106	0.105	0.101
log10(Energy_Flux100)	0.112	0.089	0.092	0.092
sin(GLAT)	0.055	0.081	0.087	0.087
LP_SigCurv	0.164	0.097	0.082	0.081
cos(GLON)	0.052	0.056	0.058	0.059
sin(GLON)	0.051	0.054	0.057	0.055

Feature	Depth 1	Depth 2	Depth 3	Depth 4
LP_index1000MeV	0.159	0.167	0.160	0.163
log10(Signif_Avg)	0.077	0.128	0.140	0.142
log10(Variability_Index)	0.100	0.114	0.114	0.113
log10(Unc_Energy_Flux100)	0.128	0.109	0.105	0.108
LP_beta	0.103	0.106	0.105	0.101
log10(Energy_Flux100)	0.112	0.089	0.092	0.092
sin(GLAT)	0.055	0.081	0.087	0.087
LP_SigCurv	0.164	0.097	0.082	0.081
cos(GLON)	0.052	0.056	0.058	0.059
sin(GLON)	0.051	0.054	0.057	0.055

This Feature Is Available To Subscribers Only