Feature Importance of LightGBM

The feature importance analysis for the LightGBM model highlights the relative contribution of each feature to model predictions. Below table lists all 88 variables along with their importance, cumulative importance, and rank for LighGBM in RECLAIM.

Feature importance of all 88 variables from LightGBM.

Rank

Variable

Feature Importance (%)

Cumulative Importance (%)

1

log_OBC

4.350861

4.350861

2

AGE

3.420195

7.771056

3

LON

2.826896

10.597953

4

R_SA_cap

2.728013

13.325966

5

AECC

2.600047

15.926012

6

wind_skew

2.396463

18.322476

7

log_MAR

2.361564

20.684039

8

wind_kurt

2.315030

22.999069

9

rel_SA_mean_clip

2.128897

25.127966

10

R_shrub_bare

1.884597

27.012564

11

wind_mean

1.797348

28.809912

12

O_cv

1.779898

30.589809

13

NSSC1_std

1.762448

32.352257

14

NVGF

1.756631

34.108888

15

I_cv

1.669381

35.778269

16

NSSC1_cv

1.657748

37.436017

17

NSSC1_skew

1.628664

39.064681

18

log_LCC

1.622848

40.687529

19

ELEV

1.587948

42.275477

20

BULK

1.553048

43.828525

21

E_std

1.529781

45.358306

22

NSSC2_mean

1.512331

46.870638

23

SAND

1.494881

48.365519

24

E_mean

1.343648

49.709167

25

SLOP

1.320382

51.029549

26

wind_cv

1.314565

52.344114

27

wind_std

1.291298

53.635412

28

R_coarse_sand

1.250582

54.885993

29

log_TE

1.244765

56.130758

30

SA_cv

1.233132

57.363890

31

ASP

1.204048

58.567939

32

log_O_std

1.192415

59.760354

33

log_HGT

1.180782

60.941135

34

log_LCBS

1.151698

62.092834

35

AECI

1.134248

63.227082

36

log_ECLR

1.093532

64.320614

37

NSSC1_kurt

1.087715

65.408329

38

log_LCWB

1.081899

66.490228

39

log_SA_std

1.052815

67.543043

40

COAR

1.017915

68.560959

41

log_SOUT

1.006282

69.567241

42

NSSC1_mean

1.000465

70.567706

43

tmin_mean

1.000465

71.568171

44

log_rain_per_area

1.000465

72.568637

45

#_rain_above_50

0.977199

73.545835

46

SILT

0.959749

74.505584

47

SA_above_90

0.942299

75.447883

48

CLAY

0.936482

76.384365

49

log_LCS

0.913215

77.297580

50

tmax_mean

0.901582

78.199162

51

I_above_90

0.901582

79.100745

52

log_MAO

0.895765

79.996510

53

AECS

0.884132

80.880642

54

log_GC

0.866682

81.747324

55

I_max_persis

0.843416

82.590740

56

log_LCSG

0.820149

83.410889

57

log_ROBC

0.814332

84.225221

58

#_rain_above_10

0.791066

85.016287

59

log_DCA

0.785249

85.801536

60

#_rain_above_100

0.761982

86.563518

61

log_ESR

0.744532

87.308050

62

log_LCG

0.744532

88.052583

63

log_SA_kurt

0.721266

88.773848

64

SA_skew

0.692182

89.466031

65

log_I_std

0.692182

90.158213

66

log_FL

0.680549

90.838762

67

log_LCT

0.668916

91.507678

68

log_LCAS

0.645649

92.153327

69

log_RT

0.628199

92.781526

70

HILL

0.610749

93.392275

71

log_PAI

0.540949

93.933225

72

log_LCSV

0.535133

94.468357

73

NSSC2_max_persis

0.529316

94.997673

74

log_MAI

0.494416

95.492089

75

log_RA

0.476966

95.969055

76

R_treaa_bare

0.442066

96.411121

77

log_SIN

0.395533

96.806654

78

CURV

0.383899

97.190554

79

log_SA_mean

0.378083

97.568637

80

NSSC2_above_90

0.366450

97.935086

81

log_CA

0.360633

98.295719

82

MRB

0.331550

98.627268

83

log_SA_mean_clip

0.331550

98.958818

84

log_RP

0.308283

99.267101

85

LAT

0.302466

99.569567

86

log_LCHV

0.267566

99.837134

87

log_LCM

0.122150

99.959283

88

DLC

0.040717

100.000000