Related Experiment Video
Updated: Sep 16, 2026
Evaluating the Effect of Roadside Parking on a Dual-Direction Urban Street
Published on: January 20, 2023
Insurance claim level estimation and influencing factor analysis for electric vehicles: A case study in Changsha city
Junqing Ye1, Xin Deng1, Jinjun Tang1
1School of Traffic & Transportation Engineering, Central South University, Changsha, China.
Objectives:
Claim amount levels reflect severity risk and facilitates risk estimation in insurance practice. This study supports risk management in electric-vehicle traffic insurance by establishing an effective classification framework for claim levels.
Methods:
The Synthetic Minority Over-Sampling Technique (SMOTE) is applied to balance the distribution of claim amount levels in traffic insurance. This study applies six distinct models-Decision Tree, Random Forest, CatBoost, Support Vector Machine (SVM), Multilayer Perceptron (MLP), and Light Gradient Boosting Machine (LGBM)-to classify the "Claim Amount Level." Finally, SHapley Additive exPlanations (SHAP) method is applied to provide both local and global explanations, thereby elucidating how specific features influence the classification of "Claim Amount Level."
Results:
The claim data were collected from electric vehicles in Changsha, China. The results show that the enhanced data maintains consistency in feature distribution and improves model stability. The Random Forest model was employed to select features and estimate claim amount levels. Based on the evaluation of classification performance, class separability, and business interpretability, the Random Forest model is identified as the optimal classifier, achieving an accuracy of 0.831, a macro-average F1-score of 0.826, and a macro-average area under the curve (AUC) of 0.950. The top three most important features, based on the SHAP analysis, are Injury Claim Flag, Claim Type, and Claim Count During Policy.
Conclusions:
SHAP-based analysis reveals heterogeneous effects of key predictors. Driver behavioral faults (negligence and insufficient safety distance) interact with the injury status to further elevate the model's predicted probability within the high claim level context. These findings enhance the interpretability of machine learning models and provide insights into risk differentiation in electric vehicle insurance.
