Using Multidimensional Data to Analyze Freeway Real-Time Traffic Crash Precursors Based on XGBoost-SHAP Algorithm

Table 5

Candidate variables of traffic flow.

Variable name	Variable description
up_q	Average upstream detector volume 10−5 minutes before the crash
up_v	Average upstream detector speed 10−5 minutes before the crash
up_o	Average upstream detector occupancy 10−5 minutes before the crash
up_dif_q	The average of the absolute value of the difference in traffic flow in the adjacent lane of the upstream detector 10−5 minutes before the crash
up_dif_v	Average of the absolute value of the speed difference between adjacent lanes of the upstream detector 10−5 minutes before the crash
up_dif_o	Average of the absolute value of the difference in adjacent lane occupancy of upstream detectors 10−5 minutes before the crash
down_q	Average downstream detector volume 10−5 minutes before the crash
down_v	Average downstream detector speed 10−5 minutes before the crash
down_o	Average downstream detector occupancy 10−5 minutes before the crash
down_dif_q	The average of the absolute value of the difference in the volume of the adjacent lanes of the downstream detector 10−5 minutes before the crash
down_dif_v	Average of the absolute value of the speed difference between adjacent lanes of the downstream detector 10−5 minutes before the crash
down_dif_o	Average of the absolute value of the difference in adjacent lane occupancy of downstream detectors 10−5 minutes before the crash
dif_q	Absolute value of the difference in volume between the upstream and downstream detectors 10−5 minutes before the crash
dif_v	Absolute value of the speed difference between upstream and downstream detectors 10−5 minutes before the crash
dif_o	Absolute value of the difference between upstream and downstream detector occupancy 10−5 minutes before the crash