Analysis of influencing parameters on children’s crossing speeds at non-signalized crosswalks using neural network models

Children-pedestrians represent a vulnerable group of road users due to the lack of protection (as other pedestrians) and additionally due to limitations related to their age, i.e., cognitive development, which significantly affects their traffic behavior. Commonly, the planning and design of urban transport infrastructure do not follow the specific needs of children-pedestrians, which is most evident in the places where they are most endangered - pedestrian crosswalks. This paper aims to analyze the parameters that affect the behavior of children-pedestrians at non-signalized pedestrian crosswalks through the analysis of walking speed. An experimental part of the research was done on 11 crosswalks situated on the roads near elementary schools in two cities in Croatia. 36 parameters that potentially influence children's behavior were analyzed - parameters related to children's behavior, driver's behavior, traffic, and traffic infrastructure features. The goal is to detect the most influential parameters using different models based on neural networks. An interesting result of the analysis of different models is that stronger correlations can reduce or nullify the significance of weaker correlations, so analyses were made on the entire database and on databases in which some groups of input parameters were omitted to detect weaker correlations. The analysis of influential parameters using models based on neural networks was compared with a statistical analysis of the correlation between the input parameters and the dependent variable, children's crossing speeds in the conflict zone.