Driver risk score based on telematics data for insurance pricing

“Creating a model to predict probability of accident using continuously streaming telematics data”

The Client is a leading auto insurer in the USA.


  • Create a customized driver risk score using data from OBD and smart phone
  • Create a big data environment that can handle the processing load


  • More than 11,000 featuresdeveloped to capture driving behavior
  • External datafor speed limit extracted at every 100 ft interval
  • Multiple machine learning methodologies for building the model
  • Additional models to create indicators from App data that replicate the OBD device
  • Ported the model developed using OBD data to App data


  • ~10% higher forecast accuracy
  • Optimized Big Data environment for complex future analytics
  • Enabling data engineering and machine learning capability