Application of Neural Networks to Analyse the Spatial Distribution of Bicycle Traffic Before, During and After the Closure of the Mill Road Bridge in Cambridgeshire, United Kingdom

Amin, Shohel · 2025 · DOAJ

DOI: 10.3390/s25103225

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study investigates the spatial distribution of bicycle traffic in response to nonrecurrent congestion events, specifically the eight-week closure of the Mill Road Bridge in Cambridge, United Kingdom, in 2019. The research addresses the challenge of monitoring traffic disruptions caused by infrastructure maintenance, which often leads to unpredictable delays and safety risks for cyclists due to mixed traffic environments. While traditional intrusive sensors are costly and disruptive to install, the paper proposes using wireless sensor data coupled with Artificial Neural Networks (ANNs) to analyze traffic patterns cost-effectively. The motivation is to understand how cyclists adapt to road closures and to identify factors influencing their route choices, thereby improving traffic management and safety strategies. The methodology employs ANN models integrated with a Generalised Delta Rule (GDR) algorithm to process hourly traffic data collected from 15 VivaCity wireless sensors installed around the bridge. The dataset covers the period from June 2019 to December 2020, allowing for analysis before, during, and after the bridge closure. The ANN architecture includes input layers representing variables such as sensor proximity to the bridge, rush hour status, day of the week, and volumes of various vehicle types (cars, buses, motorbikes, large rigid vehicles, and pedestrians). The network features two hidden layers with nine and seven neurons, respectively, using sigmoid activation functions. The data was partitioned into training, testing, and validation sets to prevent overfitting, with the GDR algorithm optimizing weight adjustments through gradient descent to minimize error rates. The results reveal distinct factors influencing bicycle traffic distribution across the three phases. Before the closure, motorbike volume (44%), bus volume (34%), and proximity to the bridge (39%) were the most significant determinants of bicycle traffic. During the closure, proximity to the bridge became the dominant factor (99%), followed by the volume of large rigid vehicles (51%), indicating that cyclists were forced into unsafe detours near heavy traffic. After the bridge reopened, unclear signage continued to impact traffic patterns, with motorbike volume (17%) and large vehicle volume (24%) remaining significant factors in spatial distribution. These findings highlight that the presence of heavy vehicles and proximity to the disruption site critically shape cyclist behavior, often pushing them into hazardous mixed-traffic environments. The significance of this study lies in demonstrating the efficacy of combining wireless sensor networks with machine learning for real-time traffic analysis. It underscores the safety risks cyclists face during infrastructure closures, particularly when rerouted into areas with high heavy goods vehicle traffic. The paper concludes that cost-effective, nonintrusive sensor-based monitoring, analyzed through robust neural network models, provides valuable insights for urban planning. These insights can inform better mitigation strategies, such as improved signage, temporary infrastructure, and adaptive traffic control, to enhance cyclist safety and reduce congestion impacts during future roadworks.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• cyclist safety