پیش‌بینی نرخ جریان ورودی‌های یک‌خطه و دوخطه میدان با درنظر گرفتن اثر عابرپیاده با استفاده از شبکه عصبی مصنوعی

in response to the continuous growth of urban areas and the challenges of modern traffic management, roundabouts have gained prominence as efficient alternatives to traditional signalized intersections, offering considerable advantages in improving safety, reducing vehicular delay, minimizing conflict points, and enhancing the continuity and efficiency of traffic flow. However, accurately estimating the entry capacity of roundabouts remains a challenging and complex issue, particularly in urban environments where pedestrian activity is frequent, diverse, and often unpredictable. Pedestrians significantly influence the operational performance of roundabouts through their presence, crossing behavior, decision-making patterns, and interactions with vehicles, which may lead to reductions in entry flow, increased queuing, and variability in driver responses. Despite the importance of these effects, many traditional capacity estimation methods have primarily focused on geometric, traffic, and vehicular parameters while neglecting pedestrian-related factors, often resulting in unrealistic or overly optimistic assessments of performance. To overcome these limitations, this study proposes a comprehensive and intelligent modeling framework for predicting entry flow rates of single-lane and two-lane roundabouts by explicitly incorporating pedestrian effects into the analysis. Field observations were conducted at four roundabouts in Mashhad, Iran, each representing different geometric layouts, traffic demands, land-use contexts, and pedestrian conditions. Key variables extracted from the collected data included pedestrian crosswalk occupancy rate, proportion of vehicle–pedestrian interactions, frequency of pedestrian group crossings, and magnitude of vehicle speed reduction during pedestrian encounters. These behavioral and spatial factors were integrated with conventional traffic parameters, such as circulating flow, to capture complex and nonlinear interactions between vehicles and pedestrians. A multilayer perceptron (MLP) artificial neural network was developed and trained using multiple structures in MATLAB to model these relationships. Model performance was evaluated using the correlation coefficient (R) and mean squared error (MSE). The optimal model achieved a high correlation coefficient (R = 0.89) and a low MSE, clearly outperforming the classical multiple linear regression (R = 0.76), thereby demonstrating the superior capability of neural networks in representing behavioral interactions that conventional linear models fail to capture. Sensitivity analysis further revealed that pedestrian crosswalk occupancy rate and vehicle speed reduction during pedestrian interactions were the most influential factors affecting roundabout entry capacity, followed by circulating flow. Excluding pedestrian-related variables led to a marked decrease in predictive accuracy, underscoring their essential role in capacity estimation and performance evaluation. These findings highlight the need for transportation engineers and planners to consider both spatial and behavioral pedestrian parameters in roundabout design, capacity modeling, and operational analysis. Neglecting these factors may result in overestimation of capacity and underestimation of delay, particularly in dense urban areas with significant pedestrian volumes and mixed traffic conditions. The proposed MLP-based model provides a robust and intelligent tool for predicting roundabout entry flow under realistic conditions, offering a more accurate and data-driven approach for traffic performance assessment. By integrating pedestrian effects into analytical frameworks, this study supports informed decision-making in the planning, design, and optimization of sustainable and safe urban transportation systems, ultimately contributing to improved operational efficiency, network reliability, and user satisfaction across all travel modes.