Award Date

May 2025

Degree Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Civil and Environmental Engineering and Construction

First Committee Member

Hualiang Teng

Second Committee Member

Moses Karakouzian

Third Committee Member

Samaan Ladkany

Fourth Committee Member

Ashok Singh

Fifth Committee Member

Ian McDonough

Number of Pages

275

Abstract

Traffic incidents have a significant impact on freeway operations, leading to severe delays, congestion, fuel wastage, and economic losses for commuters. Each year, billions of gallons of fuel are wasted, and drivers incur thousands of dollars in lost time due to incident-related delays. To mitigate these effects, Traffic Management Centers (TMCs) implement incident management strategies aimed at reducing both the frequency and severity of incidents while ensuring their prompt and safe clearance.

Dynamic Message Signs (DMSs) are key tools used by TMCs to communicate travel times and incident information to commuters. Under normal conditions, default travel times are displayed on DMSs. However, when an incident occurs, these messages are often replaced with generic warnings that may be vague, difficult to interpret, and provide little actionable information. While displaying estimated travel times could help drivers make more informed route decisions, TMCs currently lack the capability to predict Incident Duration and estimate incident-induced delays in real time. This limitation makes it challenging to provide drivers with accurate travel time updates precisely when they are needed most.

To address this challenge, this study focuses on three key objectives:1. Collecting and processing reliable traffic data related to incidents and freeway conditions. 2. Developing machine learning models to predict Incident Duration accurately. 3. Estimating real-time delays and integrating them into DMS messages to assist drivers before reaching an incident.

Multiple datasets are utilized to achieve the study's objectives. The first key source is the Incident Database (IDB) provided by FAST, which contains records of all reported incidents on the Las Vegas freeway system. This study focuses on incidents occurring in both directions of I-15 from St. Rose Parkway to the Las Vegas Motor Speedway between September 1, 2014, and August 31, 2015, from 5:00 AM to 8:00 PM, with their impacts measured until 10:00 PM.

Another crucial dataset, the One-Minute Traffic Characteristics Database (OMDB), also provided by FAST, contains traffic data recorded at one-minute intervals for Nevada’s freeway system. This extensive database comprises 525,600 XML files, each containing 17 columns and approximately 1,600 rows, totaling 14.29 billion data points. Big data analytics plays a pivotal role in this research, utilizing statistical analysis techniques such as clustering and regression to identify patterns, trends, and correlations within the data.

As part of this study, a Video Snapshots Dataset (VSDS) was created using 15-second video snapshots of incidents. This dataset visually documents incident characteristics for 272 of the 643 recorded incidents. Based on these observations, several analyses were conducted, including calculating total blockage duration, average blockage duration, and other key incident attributes.

To further analyze incident impacts, Incident Impact Heat Map Datasets (IHDS) were generated to assess both the spatial and temporal extent of each incident. By combining these two dimensions, each incident's impact was represented as a "box", covering the affected time and location. The IHDS provides a comprehensive visualization of how incidents influenced traffic conditions across different locations and time periods throughout the study year.

Using the IHDS, impact boxes were projected for all incidents recorded in the IDB, enabling a direct comparison between incidents and traffic conditions captured in the OMDB. This process resulted in the creation of the Incident Condition Dataset (ICDS) at one-minute intervals, providing a detailed and time-specific representation of incident-affected traffic conditions. Additionally, a Non-Incident Condition Dataset (NICDS) was generated to serve as a baseline for normal traffic conditions. The NICDS enables comparisons between incident-induced disruptions and typical traffic flow, improving the accuracy of delay estimations and impact analyses.

After processing, sorting, and restructuring the data, Incident Duration was modeled using machine learning methods. Three models were tested:1. Multiple Linear Regression 2. Lasso Regression with 10-fold Cross-Validation 3. Ridge Regression with 10-fold Cross-Validation

Among these, Ridge Regression demonstrated the best performance in predicting Incident Duration and was selected as the final model.

After Incident Duration predictions were in place, delay estimations were developed to support real-time incident management. This study introduces the Real-Time Incident Delay Estimation (RIDE) methodology, a dynamic, computationally efficient approach that estimates delay in 10-minute intervals—or less—to provide real-time insights into incident impacts.

Unlike traditional methods that rely on static Total Delay calculations, RIDE continuously updates Incident Duration predictions using machine learning techniques and recalculates delay estimates based on evolving conditions at the incident scene. The methodology incorporates:• Time-Interval Delay (TID): Capturing incremental delays across multiple time segments. • Average Time-Interval Delay (AvgTID): Quantifying the per-vehicle delay within each time interval.

This study also introduces a Ratio-Based Approach (RBA) to dynamically distribute Total Delays while addressing real-world complexities such as lane closures and reopenings, responder activities, and fluctuating traffic conditions.

Results indicate that RIDE is computationally efficient, replicable by Traffic Management Centers, and enables Dynamic Message Signs to display real-time, high-precision delay estimates, ultimately providing drivers with actionable travel time information.

Through the integration of granular delay estimations, spatial-temporal impact analysis, and real-time adaptability, this research provides a robust, data-driven framework for freeway incident management, helping to reduce congestion, improve traffic flow, and enhance traveler decision-making.

Keywords

Additional Travel Time; Big Data Analytics; Incident Duration; Ratio-Based Approach (RBA); Real-Time Incident Delay (RIDE); Traffic Management Center (TMC)

Disciplines

Civil Engineering

Degree Grantor

University of Nevada, Las Vegas

Language

English

Rights

IN COPYRIGHT. For more information about this rights statement, please visit http://rightsstatements.org/vocab/InC/1.0/

Download


Share

COinS