Analysis and modeling of violations crossing behavior of elderly pedestrians at signalized intersections

As the population ages, traffic violations by elderly pedestrians at signalized intersections and related issues have become increasingly prominent. Given that signalized intersections are areas where traffic conflicts are highly concentrated, their safety and operational conditions have drawn significant attention. An in-depth examination of the different types of traffic violations committed by older pedestrians and the variations in their causes will help provide a clear understanding of the characteristics of older pedestrians' crossing behavior. Therefore, the study aims to develop an interpretable predictive analytics framework for time- and space-related traffic violations committed by elderly pedestrians at signalized intersections, identify the factors associated with these two types of violations and their relative contributions to the traffic violations predictive model, thereby providing guidance for improving pedestrian safety for the old people and for age-friendly design at signalized intersections.

Elderly pedestrians' crossing violations were classified into temporal violations and spatial violations according to relevant standards. A total of 29 potential influencing factors were identified across three categories: individual characteristics, traffic conditions, and intersection facilities. Based on questionnaire surveys, field observations, and video recordings, 395 valid crossing-event samples were obtained. Firstly, considering the risks of overfitting and model instability caused by a relatively large number of candidate variables, the LASSO logistic regression was applied to screen the candidate variables. Subsequently, the variables retained after LASSO screening were incorporated into multivariate logistic regression models to estimate the relevance between each factor and elderly pedestrians' temporal violations or spatial violations, with odds ratios (ORs) and 95% confidence intervals reported. On this basis, XGBoost models were developed to predict temporal violations and spatial violations, with random forest and Naive Bayes models used as benchmark models. Finally, the SHAP method was applied to interpret the contribution of each variable to the prediction results of the XGBoost models.

The results indicate that the factors associated with temporal and spatial violations differed. For temporal violations, the multivariate logistic regression results showed that pension, crossing frequency, waiting time, green light duration, crosswalk length, bus stop near crosswalk, and right-turn signal control showed relatively strong statistical evidence of association. The SHAP results further indicated that waiting time, crossing frequency, and pension made the largest contributions to the XGBoost prediction of temporal violations. For spatial violations, educational attainment, crossing frequency, optimism bias, and crosswalk length showed relatively strong statistical evidence of association, while optimism bias, purpose of trip, and crosswalk length contributed most to the XGBoost prediction of spatial violations.

Traffic violations by elderly pedestrians when crossing the street are influenced by a combination of factors, including individual cognitive and psychological characteristics, intersection infrastructure, and traffic conditions. The study classifies elderly pedestrians' crossing violations into two categories: temporal violations and spatial violations. The associations between the selected factors and crossing violations were estimated and analyzed. In addition, the predictive contributions of these factors were interpreted using the SHAP method. The findings indicate that measures such as improving traffic safety awareness among elderly pedestrians, strengthening traffic safety education, enhancing safety facilities at intersections, and implementing targeted training may help reduce crossing violations among elderly pedestrians. Meanwhile, the findings can provide theoretical support for developing measures to improve crossing safety for elderly pedestrians and promoting age-friendly traffic design. However, it should be noted that although such optimization measures may improve crossing convenience for elderly pedestrians, they may also affect intersection operational efficiency and increase travel times for other road users. Therefore, further research is needed to balance the crossing needs of elderly pedestrians with the operational efficiency of intersections.