Idea
A predictive modeling platform integrating crash narratives and structured data to improve injury severity forecasts for transportation safety agencies
Research Paper
Core Innovation
This paper uniquely integrates unstructured police crash narratives with structured crash data using advanced NLP techniques and addresses class imbalance with KNN-based oversampling. It also incorporates roadway classification schemes to capture roadway heterogeneity, improving injury severity prediction accuracy beyond prior models that used only structured data or simpler text features.
Market Size (TAM)
$2–10B TAM, $1–2B SAM; assumption: global transportation safety and insurance sectors require advanced crash severity prediction tools.
Potential Customers & Pain Points
- Transportation Safety Agencies needing accurate injury severity predictions
- Insurance Companies seeking better risk assessment models
- Urban Planners requiring detailed crash analysis
- Law Enforcement Agencies aiming to improve crash reporting and response
- Traffic Data Analysts facing unstructured narrative data challenges
Business Model
Subscription-based SaaS platform offering predictive analytics APIs and customized injury severity modeling solutions for transportation and insurance sectors
Competitive Landscape
- RapidSOS
- Cognata
- Waycare
Implementation Challenges
- Data privacy and access restrictions for police narratives
- Integration complexity with existing traffic safety systems
- Model generalization across diverse roadway environments
Validation Strategy
- Pilot deployment with a state transportation agency
- Benchmarking against existing severity prediction models
- User feedback collection from safety analysts and insurers
Research Paper Overview
Predicting person-level injury severity using crash narratives: A balanced approach with roadway classification and natural language process techniques
Summary
This study improves traffic crash injury severity prediction by combining unstructured police crash narratives with structured data using NLP methods TF-IDF and Word2Vec. It addresses class imbalance through KNN-based oversampling and incorporates roadway classification schemes to handle heterogeneity. Evaluating 102 machine learning models with ensemble algorithms shows narrative data enhances accuracy, with TF-IDF and XGBoost performing best. The framework supports transportation safety professionals in better crash severity modeling and policy design.