Back to search

Rowan University

A Comprehensive ML and AI Framework for Intersection Safety: Assessing Contributing Factors, Surrogate Safety Measures, Non-Compliance Behaviors, and Cost-Inclusive Methodology

Abstract

dc:description.abstract

Intersection safety, particularly in traffic management, has emerged as a pressing concern due to its substantial contribution to road crashes and fatalities. In recent years, understanding road users' behavior and conflicts at intersections has become essential for evaluating traffic safety. According to the Federal Highway Administration (FHWA), in 2020, over 50% of fatal and injury crashes occurred at or near intersections, necessitating further investigation. This study addresses this issue through an innovative blend of comprehensive literature review, advanced machine learning algorithms, artificial intelligence (AI) technologies, surrogate safety measures (SSMs), non-compliance behavior, and emphasizing the economic aspects of traffic incidents. First, the study conducts a detailed analysis of intersection-related crashes in New Jersey over five years, utilizing cutting-edge machine learning techniques such as RandomForest, XGBoost, LightGBM, CatBoost, and Ensemble models, along with Shapely Additive Explanations (SHAP) impact value techniques. These methods are employed to pinpoint critical factors contributing to crash severity. Second, it explores SSMs like Time-to-Collision (TTC) and Post-Encroachment Time (PET), employing various data collection methods, including videography, LiDAR, and GPS tracking. The effectiveness of these measures is assessed using Extreme Value Theory (EVT)-based models to predict crash probabilities, aiming to improve safety while understanding the financial implications of crashes. Third, a significant innovation is the development of an AI-based video analytic tool integrating advanced detection models like YOLO-v5 and DeepSORT algorithms. This tool enables real-time safety evaluation, i.e., SSMs (e.g., TTC, PET) for vehicle-to-vehicle and pedestrian-to-vehicle conflicts, and non-compliance behaviors – such as pedestrians walking outside the crosswalk and vehicles running red lights, aligning safety metrics with cost estimates. Fourth, the study analyzes comprehensive data from Chicago to establish a correlation between red-light violations and crashes, quantifying their economic burden and guiding the prioritization of safety interventions based on their cost-effectiveness. Fifth, this study evaluates the precision and applicability of trajectory data from video analytics, GPS tracking devices, and On-Board Units (OBUs) to assess the efficacy of Connected Vehicle (CV)-related safety applications. In conclusion, this study proposes actionable, cost-inclusive recommendations for enhancing intersection safety. It advocates for a comprehensive approach that blends traditional traffic safety measures with AI and machine learning techniques, maintaining a cost-conscious perspective. This methodology not only deepens the understanding of intersection dynamics but also fosters the creation of effective, economically viable, and data-driven safety interventions. This represents a significant step forward in reducing road fatalities and injuries and managing the economic impact of traffic crashes. As the first of its kind, this study can provide transportation agencies with valuable information about intersection safety from different standpoints, including data collection, analysis, economic impact, and safety countermeasures.

Degree

thesis:*
Name thesis:degree_name
Doctor of Philosophy (Ph.D.)
Level thesis:degree_level
Dissertation
Discipline thesis:degree_discipline
Civil and Environmental Engineering

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Patel, Deep
Contributors dc:contributor
  • Mohammad Jalayer, Ph.D.
  • Yusuf Mehta, Ph.D., P.E.
  • Nidhal Carla Bouaynaya, Ph.D.

Subjects

dc:subject × 8

Identifiers

dc:identifier.*
Repository record dc:identifier
https://rdw.rowan.edu/etd/3305

Chain of custody

source
Harvested from
Rowan University
Base URL
rdw.rowan.edu/do/oai/
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Patel, Deep. A Comprehensive ML and AI Framework for Intersection Safety: Assessing Contributing Factors, Surrogate Safety Measures, Non-Compliance Behaviors, and Cost-Inclusive Methodology. Dissertation thesis, https://rdw.rowan.edu/etd/3305