Safety SDK - Traffic Crash Predictions and Data Modeling
Urban SDK is excited to announce the release of Safety SDK! As apart of our release 0.3, you can access our predictive safety algorithm and performance measurement tools for law enforcement, first responders, and insurance agencies. Over the last 4 months, our team developed a Safety SDK in partnership with Florida Highway Patrol (FHP) to help them achieve:
- Staffing Prioritization: Dispatch a required number of Troopers to specific patrolling zones with high probability of incidents by time.
- Strategic Positioning: Allocate Troopers to high probability locations of predicted incidents in order to decrease response times.
- Performance Measurement - Track key performance indicators of incident response times to prioritize staffing, positioning, and overall efficiency for Florida Highway Patrol.
The feature is available today for all smart mobility platform customers.
What We Built
Urban SDK created an application for highway patrol, law enforcement, and first responders with the objective to predict traffic incidents by location and time. Our Safety SDK is used by both supervisors and field staff to predict staffing, dispatching, and organizational needs. This product provides prescriptive analytics and data visualization for intelligent allocation of staff resources and positioning to decrease incident response and clearance times.
How We Built
Working with the Florida Highway Patrol, we connected an extensive data set of historical roadway incidents along with attributes related to each incident to real-time and near real-time roadway conditions. We modified our application to segment all roadways throughout Florida. In order to enhance the model, additional data sources were included to identify features contributing to incidents. This process can be translated across various Urban SDK solutions. This project required three components, data engineering, predictive modeling, and forecasting of future events.
Using historical incident records, we trained the model using negative samples to differentiate between incidents and non incidents. Using our proprietary tools, we generated predictions through features of road segments. From the National Oceanic and Atmospheric Administration (NOAA) we included weather data for the location and time of the incident. With a Rare Events Logistic Regression Model, we are now predicting traffic incidents with 77% accuracy based on historical comparables.
Check out the product overview and learn about predictive analytics for law enforcement agencies.