Enterprise AI Analysis
Predicting Crash Severity using Naturalistic Driving Data and Neural Networks
This study leverages AI, specifically Feedforward Neural Networks (FNNs) and SHAP analysis, with Naturalistic Driving Data (NDD) from the SHRP-2 dataset to accurately predict crash severity. Key findings include the high predictive performance of FNNs (over 93% accuracy for severe crashes), the effectiveness of SMOTE in addressing class imbalance, and the identification of crucial crash predictors like near-miss events and driver responsibility indicators. The research provides a transparent, interpretable framework for understanding real-world driver behavior and emphasizes targeted interventions and integration of AI in traffic safety management.
Executive Impact & Key Metrics
Our AI-driven analysis reveals significant advancements in crash severity prediction, offering tangible benefits for enhancing road safety and operational efficiency.
0
Severe Crash Prediction Accuracy
0
Moderate Crash Prediction Accuracy
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
Study Methodology Overview
SHRP-2 Dataset (2800x56)
→
SMOTE (Balance Classes)
→
Class Labeling (Severe/Moderate)
→
Remove NAs & Standardize Features
→
Logistic Regression (Feature Selection)
→
SHAP Analysis (Model Interpretation)
→
Neural Network (Prediction)
→
Model Evaluation (Metrics)
| Target Variables | Class 0 (Before SMOTE) | Class 1 (Before SMOTE) | Class 2 (Before SMOTE) | Total (Before SMOTE) | Class 0 (After SMOTE) | Class 1 (After SMOTE) | Class 2 (After SMOTE) | Total (After SMOTE) |
|---|---|---|---|---|---|---|---|---|
| Severity1Crash_six | 2771 | 29 | NA | 2800 | 2722 | 2722 | NA | 5444 |
| Severity2Crash_six | 2740 | 58 | 2 | 2800 | 2691 | 2691 | 2691 | 8073 |
SMOTE (Synthetic Minority Oversampling Technique) effectively addressed class imbalance, creating synthetic samples for minority classes to enhance predictive performance without compromising feature distributions.
| Model | Class | Accuracy (%) | Precision (%) | Recall (%) | F1-score (%) |
|---|---|---|---|---|---|
| Model 1 for severe crashes | 0 | 88.59 | 98.03 | 88.59 | 93.07 |
| 1 | 98.11 | 89.00 | 98.11 | 93.33 | |
| Model 2 for moderate severity crashes | 0 | 85.35 | 99.79 | 85.35 | 92.00 |
| 1 | 99.81 | 86.79 | 99.81 | 92.84 | |
| 2 | 100.00 | 100.00 | 100.00 | 100.00 |
FNN models demonstrated strong performance, with Model 1 for severe crashes achieving 93.33% F1-score for Class 1, and Model 2 for moderate crashes achieving 100% F1-score for Class 2, showcasing the models' robust predictive power across different severity levels.
Key Predictors of Crash Severity
SHAP analysis identified 'number of crashes and near-crashes in six months' as the most critical predictor for severe crashes, followed by 'number of near-crashes' and 'presence or absence of a crash in six months'. For moderate crashes, 'presence or absence of a crash in six months' was most significant, with 'number of at-fault crashes' and 'driver behavior questionnaire (DBQ) factors' also playing crucial roles. These insights highlight the importance of monitoring near-miss events and driver responsibility.
- Near-Miss Events: Frequent near-crashes significantly increase the likelihood of severe crashes.
- Driver Responsibility: At-fault incidents directly indicate risky driving behavior and are strong predictors.
- Driving Style: Aggressive behaviors like 'hard right yaw movements' correlate with increased crash severity.
Calculate Your Potential AI-Driven Safety ROI
Estimate the operational savings and reclaimed hours your enterprise could achieve by integrating advanced AI for predictive traffic safety.
Annual Operational Savings
$0
Annual Hours Reclaimed
0
Your AI Implementation Roadmap
Our structured approach ensures a seamless integration of predictive AI, tailored to your enterprise's unique needs and infrastructure.
Phase 1: Discovery & Data Integration
Comprehensive assessment of existing data infrastructure and safety protocols. Integration of naturalistic driving data (NDD) and relevant operational datasets for baseline analysis.
Phase 2: Model Development & Training
Custom development of FNN models, applying techniques like SMOTE for class imbalance and SHAP for interpretability. Rigorous training and validation using your specific data.
Phase 3: Deployment & Monitoring
Seamless deployment of predictive models into your operational environment. Continuous monitoring of model performance and real-time crash severity predictions for immediate insights.
Phase 4: Optimization & Scalability
Ongoing model refinement, feature engineering, and integration with ADAS and traffic management systems. Scaling the solution to cover expanding fleet sizes and diverse geographical operations.
Ready to Transform Your Traffic Safety?
Leverage the power of predictive AI to mitigate crash risks, enhance driver behavior, and ensure a safer future for your operations.
Ready to Get Started?
Book Your Free Consultation.
Let's Discuss Your AI Strategy!
Lets Discuss Your Needs
Select a Date
Sun
Mon
Tue
Wed
Thu
Fri
Sat