A Gradient Boosting Crash Prediction Approach for Highway-Rail Grade Crossing Crash Analysis

Table 3

Variable importance based on the GB model.

Variable	Relative importance	Influence percent (%)	Cumulative influence
AADT_N	100.00	11.00	11.00
DAYTHRU	73.33	8.07	19.07
SPSEL	67.73	7.45	26.52
NGHTTHRU	58.28	6.41	32.93
AVERAGE_TRAIN_SPEED	54.93	6.04	38.98
TRAFICLN	45.62	5.02	44.00
HWYSYS	37.98	4.18	48.18
ADVWARN	37.19	4.09	52.27
TOTAL_NUMBER_TRACK	35.75	3.93	56.20
PAVEMRK	34.15	3.76	59.96
XANGLE	32.85	3.61	63.57
FLASHPAI	31.24	3.44	67.01
COMPOWER	27.52	3.03	70.04
HIGHWAY_STOP	27.36	3.01	73.05
NEAR_CITY	26.31	2.89	75.94
XBUCK	25.73	2.83	78.77
HIGHWAY_PAVED	21.93	2.41	81.19
FLASHMA	21.42	2.36	83.54
SCHLBUS	19.78	2.18	85.72
SGNLEQP	19.45	2.14	87.86
GATES	18.00	1.98	89.84
TRUCKLN	16.70	1.84	91.68
STOPSTD	14.24	1.57	93.24
DOWNST	13.58	1.49	94.74
WHISTBAN	13.46	1.48	96.22
DAYSWT	13.19	1.45	97.67
NGHTSWT	11.80	1.30	98.97
FLASHOV	9.37	1.03	100.00
FLASHNOV	0.00	0.00	100.00
WIGWAGS	0.00	0.00	100.00