Validation of a smartphone telematics algorithm for classifying driver trips.

Area of Science:

• * Mobile Sensing and Telematics
• * Transportation Safety and Insurance Technology

Background:

• * Accurate classification of vehicle trips as driver or non-driver is crucial for telematics applications, particularly in usage-based auto insurance.
• * Existing telematics algorithms rely on smartphone data to infer driving behavior, but their accuracy needs continuous assessment.

Purpose of the Study:

• * To evaluate the accuracy of a smartphone telematics algorithm in classifying car trips as either driver or non-driver.
• * To assess the algorithm's performance across various trip types and user contexts.

Main Methods:

• * A 4-week study using the 'Way to Drive' research telematics application, mirroring algorithms used by major auto insurers.
• * Participants reviewed and self-reported misclassified trips weekly via an in-app survey.
• * Analysis focused on overall accuracy, sensitivity (driver trip classification), and specificity (non-driver trip classification).

Main Results:

• * High overall accuracy (96.5%) in distinguishing driver from non-driver trips.
• * Excellent sensitivity (97.5%) for correctly identifying driver trips.
• * Slightly lower but acceptable specificity (91.2%) for non-driver trips, with notable variability.

Conclusions:

• * The smartphone telematics algorithm demonstrates high accuracy for driver vs. non-driver trip classification.
• * The algorithm's performance is robust across diverse phone, vehicle, and driving conditions.
• * This technology shows significant promise for accurate data collection in telematics and auto insurance contexts.