US 12,393,645 B2
System and/or method for personalized driver classifications
Sambuddha Bhattacharya, San Francisco, CA (US); Nikhil Mudaliar, San Francisco, CA (US); Rajesh Bhat, San Francisco, CA (US); Udit Gupta, San Francisco, CA (US); and Tapan Bhardwaj, San Francisco, CA (US)
Assigned to Credit Karma, LLC, Oakland, CA (US)
Filed by Credit Karma, LLC, Oakland, CA (US)
Filed on Dec. 2, 2022, as Appl. No. 18/073,959.
Claims priority of provisional application 63/285,251, filed on Dec. 2, 2021.
Prior Publication US 2023/0177121 A1, Jun. 8, 2023
Int. Cl. G06N 20/00 (2019.01); G06F 18/23 (2023.01); G06F 18/24 (2023.01)
CPC G06F 18/24 (2023.01) [G06F 18/23 (2023.01); G06N 20/00 (2019.01)] 20 Claims
OG exemplary drawing
 
1. A method for mobile driver-versus-passenger classification, comprising:
collecting coarse sensor data with a mobile device associated with a user;
detecting a first vehicle trip based on the coarse sensor data;
extracting a first set of features from the coarse sensor data for the first vehicle trip;
providing the extracted first set of features to a tree-based model to generate a first trip classification during the first vehicle trip, wherein:
the tree-based model comprises a personalized driver-versus-passenger (DvP) model and a global DvP model, and
the personalized DvP model comprises a hierarchical clustering model for the user of the mobile device, trained based on historical vehicular trips completed by the user;
in response to the first trip classification, triggering an action at the mobile device;
determining a second set of features from sensor data collected with the mobile device during a second vehicular trip;
based on the second set of features, disqualifying the personalized DvP model;
in response to disqualify the personalized DvP model, providing the second set of features to the global DvP model to generate a second trip classification during the second vehicular trip, wherein the global DvP model is operable to generate the second trip classification in the event the personalized DvP model has been disqualified, thereby enabling the second trip classification to be produced; and
retraining the personalized DvP model with the second set of features and the second trip classification, thereby increasing an alignment of the personalized DvP model with a driving history of the user.