Individual Driver Crash Risk Classification Based on IoV Data and Offline Consumer Behavior Data
Table 9
Comparison of different consumer behavior-based models for collision-free and collision risk classification (binning data).
Dependent variable
LRbox
NNbox
RFbox
SVMbox
AUC
IR (%)
AUC
IR (%)
AUC
IR (%)
AUC
IR (%)
Basic features
0.747
—
0.752
—
0.722
—
0.690
—
Basic features + proportion of maintenance carried out in 4S dealer shops
0.771
3.23
0.768
2.14
0.750
3.92
0.729
5.56
Basic features + average price of maintenance carried out in 4S dealer shops
0.751
0.55
0.750
−0.25
0.735
1.77
0.718
4.01
Basic features + average maintenance interval (days)
0.829
10.92
0.811
7.87
0.808
11.92
0.772
11.75
Basic features + average maintenance interval (miles)
0.765
2.40
0.755
0.41
0.742
2.83
0.714
3.43
Basic features + average number of products used for one maintenance episode
0.750
0.41
0.737
−1.96
0.718
−0.51
0.707
2.33
Note 1: LRbox, NNbox, RFbox, and SVMbox represent LR, NN, RF, and SVM models using binning data. Note 2: IR represents the increase in the AUC compared with the basic feature-based model.
