Figures (6)  Tables (3)

    • Figure 1. 

      Framework for fatal traffic accident prediction and causal analysis.

    • Figure 2. 

      Model structure of CatBoost and SHAP.

    • Figure 3. 

      Receiver operating characteristic curve for each compared algorithm.

    • Figure 4. 

      Contributing factors for fatal accidents: (a) feature importance, and (b) total effect.

    • Figure 5. 

      Feature values and main effects of contributing factors for fatal accidents: (a) road type (feature value), (b) road type (main effect), (c) hour (feature value), (d) hour (main effect), (e) weather (feature value), (f) weather (main effect), (g) accident cause (feature value), and (h) accident cause (main effect).

    • Figure 6. 

      Interaction effects of contributing factors for fatal accidents: (a) road administrative class vs median divider, (b) road administrative class vs roadside protection, (c) road type vs road administrative class, (d) mode of transport vs accident liability, (e) mode of transport vs road administrative level, (f) mode of travel vs accident type, (g) accident type vs road administrative class, and (h) driver age vs mode of transport.

    • Feature Value Abbreviation of value
      Age (0,20] / (20,30] / (30,40] / (40,50] / (50,60] / (60,70] / (70, ) 0-20 / 20-30 / 30-40 / 40-50 / 50-60 / 60-70 / > 70
      Gender Female / Male −
      Driving with license True / False T / F
      Driving with seatbelt True / False / Unknown T / F / U
      Drunk driving True / False T / F
      Fatigued driving True / False T / F
      Speeding True / False T / F
      Overloaded True / False T / F
      Hit-and-run driving True / False T / F
      Mode of transport Pedestrian / Bicycle / Electric bicycle / Motorcycle / Tricycle / Car / Bus / Small truck / Large truck / Combination vehicle / Other MOD1 / MOD2 / MOD3 / MOD4 / MOD5 / MOD6 / MOD7 / MOD8 / MOD9 / MOD10 / MOD11
      Road administrative class National road / Provincial road / Country road / Township road / Urban road ADM1 / ADM2 / ADM3 / ADM4 / ADM5
      Road type Road segment / Intersection / Road access / Ramp / Bridge or tunnel / Elevated road / Other RT1 / RT2 / RT3 / RT4 / RT5 / RT6 / RT7
      Traffic control No control / Marking / Sign / Traffic light / Other TC1 / TC2 / TC3 / TC4 / TC5
      Median divider No divider / Greenbelt / Pier / Guardrail CI1 / CI2 / CI3 / CI4
      Roadside protection No protection / Greenbelt / Pier / Guardrail / Other SP1 / SP2 / SP3 / SP4 / SP5
      Road condition Intact / Damaged RC1 / RC2
      Road material Asphalt / Cement / Other RM1 / RM2 / RM3
      Road surface Dry / Wet / Ponding / Icy / Other RS1 / RS2 / RS3 / RS4 / RS5
      Month January / February / March / April / May / June / July / August / September / October / November / December Jan / Feb / Mar / Apr / May / Jun / Jul / Aug / Sept / Oct / Nov / Dec
      Day of week Monday / Tuesday / Wednesday / Thursday / Friday / Saturday / Sunday Mon / Tue / Wed / Thur / Fri / Sat / Sun
      Hour 0 / 1 / 2 / 3 / 4 / 5 / 6 / 7 / 8 / 9 / 10 / 11 / 12 / 13 / 14 / 15 / 16 / 17 / 18 / 19 / 20 / 21 / 22 / 23 −
      Weather Sunny / Cloudy / Rainy or snowy / Foggy / Other W1 / W2 / W3 / W4 / W5
      Lighting conditions Night without street lamp / Night with street lamp / Nightfall / Dawn / Daytime LC1 / LC2 / LC3 / LC4 / LC5
      Visibility < 50 m / 50−100 m / 100−200 m / > 200 m VIS1 / VIS2 / VIS3 / VIS4
      Accident cause Pedestrian or non-motor vehicle violation / Distracted driving/ Natural disaster / Substandard vehicle / Driving in the wrong lane / Failure to give way according to traffic rules / Traffic light or sign violation / Failure to maintain safe distance / Illegal parking / Illegal overtaking or lane changing / Driving in the opposite direction / Other AC1 / AC2 / AC3 / AC4 / AC5 / AC6 / AC7 / AC8 / AC9 / AC10 / AC11 / AC12
      Accident type Overturned vehicle / Bumping into fixed object / Bumping into pedestrian(s) / Side-swipe / Side collision / Rear-end collision / Head-on collision / Other AF1 / AF2 / AF3 / AF4 / AF5 / AF6 / AF7 / AF8
      Accident liability Unknown / No liability / Secondary / Equal / Primary / Full AL1 / AL2 / AL3 / AL4 / AL5 / AL6
      Accident severity Injury or property damage (non-fatal) / Fatal AS0 / AS1

      Table 1. 

      Data description.

    • Algorithms Parameters Parameter range Optimal parameters
      CatBoost Learning rate lr lr $ \in $ {0.1, 0.2, 0.3, 0.5} lr = 0.1
      Number of estimators Ne Ne $ \in $ {300, 500, 1,000, 2,000} Ne = 500
      Depth of each tree d d $ \in $ {9, 10, 11, 12} d = 10
      LightGBM Learning rate lr lr $ \in $ {0.01, 0.05, 0.1, 0.2} lr = 0.1
      Number of estimators Ne Ne $ \in $ {300, 500, 1,000, 2,000} Ne = 1,000
      Maximum depth of each tree d d $ \in $ {10, 20, $ \infty $} d = $ \infty $
      Maximum tree leaves of each tree ml ml $ \in $ { 10, 31, 100, 200} ml = 100
      XGBoost Learning rate lr lr $ \in $ {0.1, 0.2, 0.3, 0.5} lr = 0.1
      Number of estimators Ne Ne $ \in $ {200, 300, 500, 1,000} Ne = 500
      maximum depth of each tree d d $ \in $ { 6, 20, 30, 50} d = 30
      Minimum decreasing value of loss function gamma gamma $ \in $ {0, 0.1, 0.2, 0.3} gamma = 0.1
      GBDT Learning rate lr lr $ \in $ {0.01, 0.05, 0.1, 0.2} lr = 0.1
      Number of estimators Ne Ne $ \in $ {200, 300, 500, 1,000} Ne = 500
      Maximum depth of each tree d d $ \in $ { 3, 10, 30, 50} d = 30
      ANN Number of hidden layer units Nunit Nunit $ \in $ {64, 128, 256, 512} Nunit = 256
      Learning rate lr lr $ \in $ {10−4, 10−3, 10−2} lr = 10−3
      MNL Penalty p p $ \in $ {none, l2} p = l2
      Inverse of regularization strength C C $ \in $ {0.1, 1, 10, 100} C = 10

      Table 2. 

      Parameter selection for each compared algorithm.

    • Algorithms Precision Recall F1-score
      CatBoost 0.912 0.942 0.927
      LightGBM 0.826 0.872 0.848
      XGBoost 0.820 0.847 0.833
      GBDT 0.808 0.852 0.829
      ANN 0.727 0.817 0.769
      MNL 0.596 0.651 0.622

      Table 3. 

      Metrics for each compared algorithm.