1 of 22

The Influence of Unsafe Human Behaviors on Nighttime Pedestrian Crash Injury Severity at Intersections

Sheikh Muhammad Usman

Graduate Research Assistant,

Department of Civil & Environmental Engineering

Image Source: Internet

2 of 22

Introduction

USA: More than 50% increase in pedestrian fatalities from 2009 to 2019 (FARS, 2019)
More than 85% of those fatalities occurred at night (FARS, 2019)
26% of fatal pedestrian crashes at intersections in the United States in 2018 (NHTSA, 2018)
From 2015 to 2019, nighttime pedestrian fatalities increased by 16.4%, with an even sharper increase of 18.4% observed at intersections

3 of 22

Introduction

How to strike the right balance between road user safety and mobility at intersections?
Longer red-light phases, reduction in speed limit, pedestrian refuge islands at intersections, consequently reduce vehicular mobility
Longer green light phase, and permitted right turn on red without yielding, highly increase the likelihood of collisions with pedestrians at intersections
USDOT’s Vision Zero Goal → Safe Systems Approach
Study Objective

Investigate how unsafe pedestrian and driver behaviors (dash/dart, alcohol impairment, non-yielding), traffic control measures, and environmental conditions influence nighttime pedestrian crash injury severity at intersections

4 of 22

Framework: PBCAT Crash (Typing) Data-police reports

PBCAT Pedestrian Crash Dataset

PBCAT Crash Typing Tool

Detailed Crash Descriptors

Focus on Crash Patterns at Intersections

Motorists not Yielding to Pedestrians, Turning Maneuvers, Pedestrians Walking along Road, Pedestrian Dash/Dart Out

5 of 22

Study Area: North Carolina, NC

Spatial Distribution of Nighttime Pedestrian Crashes at Intersections (N = 2733)

6 of 22

Descriptive Statistics of Key Variables

7 of 22

Cross Tabulation Results (Nighttime Crashes N = 2733)�Pedestrian and Driver Alcohol/Drug Impairment

Variable	Statistics	Pedestrian Crash Injury Severity					Total
Driver & Pedestrian Alcohol/Drug Use	Statistics	No Injury	Possible Injury	Minor Injury	Severe Injury	Fatal Injury	Total
Driver Alcohol Impairment	Frequency	1	18	15	31	27	92
	Row (%)	1.09	19.56	16.30	33.70	29.35	100
Pedestrian Alcohol Impairment	Frequency	16	60	71	132	80	359
	Row (%)	4.46	16.71	19.78	36.77	22.28	100
Both Alcohol Impaired	Frequency	0	1	6	8	12	27
	Row (%)	0	3.71	22.22	29.63	44.44	100
No Alcohol Impairment	Frequency	128	509	633	494	491	2255
	Row (%)	5.68	22.57	28.07	21.91	21.77	100

8 of 22

Initial Model – Combined Daytime and Nighttime Crashes

8

Ordered Logit Model

Nighttime Pedestrian Crashes -> Highly Significant and Positively Associated with higher injury severity, given a crash, compared to daytime crashes

Evidence for focusing on nighttime pedestrian crashes

9 of 22

Ordered Logit Model – Nighttime Crashes (N = 2733)

9

Crashes involving motorists not yielding to pedestrians at intersections significantly increase the likelihood of severe and fatal injuries, with a positive coefficient (0.29) and a marginal effect of 0.0273 for fatal injury. Pedestrian dash/dart-out behavior is another notable contributor to injury severity, increasing the probability of fatal injury by 0.005. Additionally, crash types involving left-turning vehicles—whether pedestrians are on a perpendicular or parallel path—are significantly associated with higher injury severity. Alcohol and drug impairment, particularly when both the driver and pedestrian are under the influence, shows a very strong association with severe crash outcomes. The coefficients for driver, pedestrian, and combined alcohol use are all significant, with the combined impairment increasing the probability of fatal injury by 0.194. , intersections with no traffic control (e.g., lacking signals or signs) are linked to more severe outcomes, with a marginal effect of 0.0042 for fatal injuries. Higher posted speed limits are positively correlated with increased injury severity. For instance, intersections with a speed limit of 50–75 mph show a 0.064 unit increase in the probability of fatal injury compared to roads with speed limits below 30 mph.

10 of 22

Ordered Logit: Variable Importance Plot

Both Driver and Pedestrian Alcohol Impairment, only driver, and only pedestrian alcohol impairment among key predictors.
High injury crash types: Pedestrian in Travel Lane, Motorists not Yielding to Pedestrians

Variable Importance based on Standardized Coefficients in Ordered Logit Model

11 of 22

XGBoost: Variable Importance Plot

Driver and Pedestrian alcohol impairment among key predictors.
High injury crash types: Pedestrian in Travel Lane, Motorists not Yielding to Pedestrians, Pedestrian Dash/Dart

12 of 22

XGBoost: SHAP Summary and Interaction Plots

SHAP interaction plot demonstrates the compounded danger of impaired pedestrian behavior in fast-moving traffic environments. �

13 of 22

Light GBM: Variable Importance Plot

Driver and Pedestrian alcohol impairment among key predictors
High injury crash types: Pedestrian in Travel Lane, Motorists not Yielding to Pedestrians, Pedestrian Dash/Dart
Findings consistent with Ordered Logit and XGBoost Models

14 of 22

Light GBM: SHAP Summary and Interaction Plots

Compounded risk of severe injury when an alcohol impaired driver also fails to yield to a pedestrian. �

15 of 22

Cat Boost: Variable Importance Plot

Driver and Pedestrian alcohol impairment among key predictors
High injury crash types: Pedestrian in Travel Lane, Motorists not Yielding to Pedestrians, Pedestrian Dash/Dart
Findings consistent with Ordered Logit, XGBoost, and Light GBM Models

16 of 22

16

Cat Boost: SHAP Summary and Interaction Plots

Compounding risk when drivers under the influence of alcohol strike pedestrians who suddenly dash or dart into the roadway

Intensifies fatal injury risk �

17 of 22

Cat Boost: Predictive Performance

Cat Boost outperforms other models in terms of classification accuracy.
Classification accuracy of 95.72% � �

18 of 22

LIMITATIONS

Study results may not be generalizable to other regions in the U.S.

Lack of information related to pedestrian conspicuity

(Clothing color, reflectivity, visibility aids)

Image Source: Internet

19 of 22

19

Conclusions

Objective

�

Pedestrian Actions		Driver Behaviors	Traffic Control	Environmental Conditions
Pedestrian in Travel Lane Pedestrian Dash/Dart out	Pedestrian Alcohol Impairment	Motorist Failed to Yield Driver Alcohol Impairment	Stop Controlled Intersections No Traffic Control	Not Lighted Intersection

Influence of unsafe pedestrian and driver behaviors (dash/dart, alcohol impairment, non-yielding), traffic control measures, and environmental conditions on nighttime pedestrian crash injury severity at intersections

Image Source: Internet

20 of 22

Practical Implications

Intensified Enforcement Strategies, i.e., Nighttime DUI checkpoints, in nightlife zones and near college campuses
AI-based intersection safety system, leveraging vehicle and pedestrian detection technologies, to issue real-time alerts to reduce conflicts.
Harness V2P communications to avoid high-risk crash types identified
Exclusively protected pedestrian signal phasing
Leading pedestrian intervals
Integrating pedestrian detection into intelligent adaptive traffic signals

Image Source: Internet

21 of 22

Study Summary

Data

Police-reported pedestrian crashes at intersections in North Carolina (2016-2022)
Coded using PBCAT for detailed descriptors of crash type, pedestrian behavior, and roadway conditions.
Dataset: 5,345 total crashes (daytime + nighttime); 2,733 nighttime crashes.

Methodology: Ordered Logit + XGBoost, Light GBM, Cat Boost

Key results

Driver and pedestrian alcohol impairment associated with increase in fatal injury probability.
Motorists failing to yield to pedestrians associated with increase in fatal injury probability.
Poor lighting at intersections associated with increase in fatal injury probability.
The AI methods (XGBoost, Light GBM, Cat Boost) outperformed the statistical ordered logit model in predictive accuracy (Cat Boost accuracy = 95.72%).

Study Contribution

Comprehensive study combining statistical and AI models with SHAP explainability to examine nighttime pedestrian crash injury severity at intersections.
Demonstrates potential for the development of AI-enabled intersection safety systems.

21

Overall goal: Investigate how unsafe pedestrian and driver behaviors, traffic control measures, and environmental conditions influence nighttime pedestrian crash injury severity at intersections

Image Source: Internet

22 of 22

Thank you!

Questions?