Prediction of User Perceptions of Signalized Intersection Level of Service Based on Fuzzy Neural Networks

Level of service (LOS) methodology incorporating user perceptions provides a measure of how well signalized intersections accommodate their users. This methodology offers new insights into signalized intersection LOS and may thus overcome the limitations of conventional delay-based methods to some extent. However, few studies on this perception-based LOS methodology have been conducted in China, where traffic conditions can be characterized by high volumes of pedestrians and bicycles. To fill the gap, this study proposes a fuzzy neural networks–based approach to predict user perceptions of signalized intersection LOS. Membership values are used to measure the perceived levels of the influential attributes that are difficult to capture accurately. A neural network containing fuzzy reasoning experiences is employed to combine the perceived attributes in order to determine LOS. To illustrate the proposed approach, a case study is performed on turning movement LOS under mixed traffic conditions. The LOS model is further calibrated and validated by using the data collected in a visualization-based survey conducted in Beijing. The numerical results indicate that the model has a favorable capability to predict user perceptions of LOS. Pearson correlation coefficients demonstrate the weak linear association between user perceptions (LOS ratings) and capacities (or volume-to-capacity ratios). Analyses of variance statistically support that users’ LOS ratings differ significantly at different levels of capacities (or volume-to-capacity ratios). The ratings depend strongly on delays. The survey results validate the delay-based method proposed by the Highway Capacity Manual to some extent.