Granted Patent B1

US 12,505,410 · App. 17/524,544 · Granted Dec 23, 2025

Inventors: Kenneth Jason Sanchez (San Francisco, CA); Gil Tamari (San Francisco, CA)

Assignee: QUANATA, LLC

G06Q10/1091G07C5/0808

Abstract

Claims (123)

1 . A computer-implemented method for classifying a vehicular trip, the method comprising:

obtaining a user classification for a vehicle operator associated with an unlabeled vehicular trip as for work or for personal use;

obtaining a first set of historic telematics data associated with a first set of historic vehicular trips during which the vehicle operator operated a vehicle for work;

obtaining a second set of historic telematics data associated with a second set of historic vehicular trips during which the vehicle operator operated the vehicle for personal use;

training a classification machine learning model to generate a trained classification machine learning model comprising a logistic regression model, to predict one or more attributes of the vehicle based on sensor data collected at least in part while the first set of historic telematics data and the second set of historic telematics data were collected, wherein the classification machine learning model is trained using hyperparameter tuning, cross-validation, and Bayesian methods to improve performance of the trained classification machine learning model;

obtaining a set of unlabeled telematics data associated with the unlabeled vehicular trip;

classifying the unlabeled vehicular trip, using the trained classification machine learning model to generate a machine classification that predicts the unlabeled vehicular trip as for work or for personal use by at least:

identifying:

based at least in part upon the first set of historic telematics data, a first set of baseline operation features associated with the first set of historic vehicular trips;

based at least in part upon the second set of historic telematics data, a second set of baseline operation features associated with the second set of historic vehicular trips; and

based at least in part upon the set of unlabeled telematics data, a set of representative operation features associated with the unlabeled vehicular trip;

wherein the first set of baseline operation features, the second set of baseline operation features, and the set of representative operation features each comprise one or more driving characteristics;

vectorizing each feature of the first set of baseline operation features, the second set of baseline operation features, and the set of representative operation features;

determining a first vector deviation and a second vector deviation, wherein:

the first vector deviation represents differences between the first set of baseline operation features, as vectorized, and the set of representative operation features, as vectorized; and

the second vector deviation represents differences between the second set of baseline operation features, as vectorized, and the set of representative operation features, as vectorized;

transforming, using the logistic regression model, the first vector deviation and the second vector deviation into a probability indicative of how probable the logistic regression model considers the unlabeled vehicular trip to be for work, wherein one of:

upon determining that the set of representative operation features deviates from the first set of baseline operation features less than from the second set of baseline operation features, the trained classification machine learning model classifies the unlabeled vehicular trip as for work; or

upon determining that the set of representative operation features deviates from the first set of baseline operation features more than from the second set of baseline operation features, the trained classification machine learning model, classifies the unlabeled vehicular trip as for personal use; and

validating the user classification for the unlabeled vehicular trip based at least in part upon the user classification and the machine classification predicted by using the trained classification machine learning model.

2 . The computer-implemented method of claim 1 , wherein validating the user classification comprises:

accepting the user classification upon determining that the user classification matches the machine classification predicted by using the trained classification machine learning model; or

rejecting the user classification upon determining that the user classification does not match the machine classification predicted by using the trained classification machine learning model.

3 . The computer-implemented method of claim 1 , wherein:

classifying the unlabeled vehicular trip comprises generating a confidence level associated with the machine classification predicted using the trained classification machine learning model; and

validating the user classification comprises accepting the user classification upon:

determining that the user classification does not match the machine classification predicted by using the trained classification machine learning model; and

determining that the confidence level is less than a confidence threshold.

4 . The computer-implemented method of claim 1 , wherein the validating the user classification comprises:

obtaining a community classification for a set of similar vehicular trips traveled by one or more similar vehicle operators, the one or more similar vehicle operators sharing a common employer, a common work region, or a common work schedule with the vehicle operator, the set of similar vehicular trips having similar travel paths as the unlabeled vehicular trip; and

accepting the user classification upon determining that the user classification matches the machine classification predicted by using the trained classification machine learning model and the community classification.

5 . The computer-implemented method of claim 1 , further comprising:

obtaining a work insurance claim submitted by the vehicle operator associated with the unlabeled vehicular trip; and

determining, based at least in part upon the machine classification, a claim decision by at least:

determining the claim decision as accept claim when the machine classification is for work; or

determining the claim decision as reject claim when the machine classification is for personal use.

6 . The computer-implemented method of claim 1 , further comprising:

obtaining a personal insurance claim submitted by the vehicle operator associated with the unlabeled vehicular trip; and

determining, based at least in part upon the machine classification, a claim decision by at least:

determining the claim decision as accept claim when the machine classification is for personal use; or

determining the claim decision as reject claim when the machine classification is for work.

7 . The computer-implemented method of claim 1 , further comprising:

upon classifying the unlabeled vehicular trip as for work, determining a set of policy modifications associated with a work insurance policy of the vehicle operator based at least in part upon the set of representative operation features; or

upon classifying the unlabeled vehicular trip as for personal use, determining a set of policy modifications associated with a personal insurance policy of the vehicle operator based at least in part upon the set of representative operation features.

8 . The computer-implemented method of claim 7 , wherein each of the set of policy modifications associated with the work insurance policy of the vehicle operator and the set of policy modifications comprises associated with the personal insurance policy of the vehicle operator comprises at least one of modify policy premium, modify a risk score, modify an operator behavioral score, modify a policy incentive, issue a reimbursement, or modify a vehicular operation challenge.

9 . The computer-implemented method of claim 1 , further comprising:

obtaining a first set of historic path conditions associated with the first set of historic vehicular trips;

obtaining a second set of historic path conditions associated with the second set of historic vehicular trips; and

obtaining a set of unlabeled path conditions associated with the unlabeled vehicular trip;

wherein:

identifying the first set of baseline operation features comprises calibrating the first set of baseline operation features based at least in part upon the first set of historic path conditions;

identifying the second set of baseline operation features comprises calibrating the second set of baseline operation features based at least in part upon the second set of historic path conditions; and

identifying the set of representative operation features comprises calibrating the set of representative operation features based at least in part upon the set of unlabeled path conditions.

10 . The computer-implemented method of claim 9 , wherein each of the first set of historic path conditions, the second set of historic path conditions, and the set of unlabeled path conditions comprises at least one of path curvature, path speed limit, average speed by travelers on path, traffic condition, weather condition, or time of day.

11 . The computer-implemented method of claim 1 , wherein:

identifying the first set of baseline operation features comprises identifying a first sequence of baseline operation features based at least in part upon the first set of historic telematics data;

identifying the second set of baseline operation features comprises identifying a second sequence of baseline operation features based at least in part upon the second set of historic telematics data; and

identifying the set of representative operation features comprises identifying a sequence of representative operation features based at least in part upon the set of unlabeled telematics data.

12 . The computer-implemented method of claim 1 , wherein the one or more driving characteristics comprise at least one of route choice, acceleration, maximum speed, average speed, braking, turning radius, following distance, lane changes, magnitude of jerk, magnitude of swerve, or distraction.

13 . A computing system for classifying a vehicular trip, the computing system comprising:

one or more processors; and

a memory storing computing instructions that, upon execution by the one or more processors, cause the computing system to perform one or more operations comprising:

obtaining a user classification for a vehicle operator associated with an unlabeled vehicular trip as for work or for personal use;

obtaining a first set of historic telematics data associated with a first set of historic vehicular trips during which the vehicle operator operated a vehicle for work;

obtaining a second set of historic telematics data associated with a second set of historic vehicular trips during which the vehicle operator operated the vehicle for personal use;

training a classification machine learning model to generate a trained classification machine learning model comprising a logistic regression model, to predict one or more attributes of the vehicle based on sensor data collected at least in part while the first set of historic telematics data and the second set of historic telematics data were collected, wherein the classification machine learning model is trained using hyperparameter tuning, cross-validation, and Bayesian methods to improve performance of the trained classification machine learning model;

obtaining a set of unlabeled telematics data associated with the unlabeled vehicular trip;

classifying the unlabeled vehicular trip, using the trained classification machine learning model to generate a machine classification that predicts the unlabeled vehicular trip as for work or for personal use by at least:

identifying:

based at least in part upon the first set of historic telematics data, a first set of baseline operation features associated with the first set of historic vehicular trips;

based at least in part upon the second set of historic telematics data, a second set of baseline operation features associated with the second set of historic vehicular trips; and

based at least in part upon the set of unlabeled telematics data, a set of representative operation features associated with the unlabeled vehicular trip;

wherein the first set of baseline operation features, the second set of baseline operation features, and the set of representative operation features each comprise one or more driving characteristics;

vectorizing each feature of the first set of baseline operation features, the second set of baseline operation features, and the set of representative operation features;

determining a first vector deviation and a second vector deviation, wherein:

the first vector deviation represents differences between the first set of baseline operation features, as vectorized, and the set of representative operation features, as vectorized; and

the second vector deviation represents differences between the second set of baseline operation features, as vectorized, and the set of representative operation features, as vectorized;

transforming, using the logistic regression model, the first vector deviation and the second vector deviation into a probability indicative of how probable the logistic regression model considers the unlabeled vehicular trip to be for work, wherein one of:

upon determining that the set of representative operation features deviates from the first set of baseline operation features less than from the second set of baseline operation features, the trained classification machine learning model classifies the unlabeled vehicular trip as for work; or

upon determining that the set of representative operation features deviates from the first set of baseline operation features more than from the second set of baseline operation features, the trained classification machine learning model classifies the unlabeled vehicular trip as for personal use; and

validating the user classification for the unlabeled vehicular trip based at least in part upon the user classification and the machine classification predicted by using the trained classification machine learning model.

14 . A non-transitory computer-readable medium storing computing instructions for classifying a vehicular trip, the computing instructions upon execution by one or more processors of a computing system, cause the computing system to perform one or more operations comprising:

obtaining a user classification for a vehicle operator associated with an unlabeled vehicular trip as for work or for personal use;

obtaining a first set of historic telematics data associated with a first set of historic vehicular trips during which the vehicle operator operated a vehicle for work;

obtaining a second set of historic telematics data associated with a second set of historic vehicular trips during which the vehicle operator operated the vehicle for personal use;

training a classification machine learning model to generate a trained classification machine learning model comprising a logistic regression model to predict one or more attributes of the vehicle based on sensor data collected at least in part while the first set of historic telematics data and the second set of historic telematics data were collected, wherein the classification machine learning model is trained using hyperparameter tuning, cross-validation, and Bayesian methods to improve performance of the trained classification machine learning model;

obtaining a set of unlabeled telematics data associated with the unlabeled vehicular trip;

classifying the unlabeled vehicular trip, using the trained classification machine learning model to generate a machine classification that predicts the unlabeled vehicular trip as for work or for personal use by at least:

identifying:

based at least in part upon the first set of historic telematics data, a first set of baseline operation features associated with the first set of historic vehicular trips;

based at least in part upon the second set of historic telematics data, a second set of baseline operation features associated with the second set of historic vehicular trips; and

based at least in part upon the set of unlabeled telematics data, a set of representative operation features associated with the unlabeled vehicular trip;

wherein the first set of baseline operation features, the second set of baseline operation features, and the set of representative operation features each comprise one or more driving characteristics;

vectorizing each feature of the first set of baseline operation features, the second set of baseline operation features, and the set of representative operation features;

determining a first vector deviation and a second vector deviation, wherein:

the first vector deviation represents differences between the first set of baseline operation features, as vectorized, and the set of representative operation features, as vectorized; and

the second vector deviation represents differences between the second set of baseline operation features, as vectorized, and the set of representative operation features, as vectorized;

transforming, using the logistic regression model, the first vector deviation and the second vector deviation into a probability indicative of how probable the logistic regression model considers the unlabeled vehicular trip to be for work, wherein one of:

upon determining that the set of representative operation features deviates from the first set of baseline operation features less than from the second set of baseline operation features, the trained classification machine learning model classifies the unlabeled vehicular trip as for work; or

upon determining that the set of representative operation features deviates from the first set of baseline operation features more than from the second set of baseline operation features, the trained classification machine learning model classifies the unlabeled vehicular trip as for personal use; and

validating the user classification for the unlabeled vehicular trip based at least in part upon the user classification and the machine classification predicted by using the trained classification machine learning model.

15 . The computing system of claim 13 , wherein validating the user classification comprises:

accepting the user classification upon determining that the user classification matches the machine classification predicted by using the trained classification machine learning model; or

rejecting the user classification upon determining that the user classification does not match the machine classification predicted by using the trained classification machine learning model.

16 . The computing system of claim 13 , wherein:

classifying the unlabeled vehicular trip comprises generating a confidence level associated with the machine classification predicted using the trained classification machine learning model; and

validating the user classification comprises accepting the user classification upon:

determining that the user classification does not match the machine classification predicted by using the trained classification machine learning model; and

determining that the confidence level is less than a confidence threshold.

17 . The computing system of claim 13 , wherein the validating the user classification comprises:

obtaining a community classification for a set of similar vehicular trips traveled by one or more similar vehicle operators, the one or more similar vehicle operators sharing a common employer, a common work region, or a common work schedule with the vehicle operator, the set of similar vehicular trips having similar travel paths as the unlabeled vehicular trip; and

accepting the user classification upon determining that the user classification matches the machine classification predicted by using the trained classification machine learning model and the community classification.

18 . The non-transitory computer-readable medium of claim 14 , wherein validating the user classification comprises:

accepting the user classification upon determining that the user classification matches the machine classification predicted by using the trained classification machine learning model; or

rejecting the user classification upon determining that the user classification does not match the machine classification predicted by using the trained classification machine learning model.

19 . The non-transitory computer-readable medium of claim 14 , wherein:

classifying the unlabeled vehicular trip comprises generating a confidence level associated with the machine classification predicted using the trained classification machine learning model; and

validating the user classification comprises accepting the user classification upon:

determining that the user classification does not match the machine classification predicted by using the trained classification machine learning model; and

determining that the confidence level is less than a confidence threshold.

20 . The non-transitory computer-readable medium of claim 14 , wherein the validating the user classification comprises:

obtaining a community classification for a set of similar vehicular trips traveled by one or more similar vehicle operators, the one or more similar vehicle operators sharing a common employer, a common work region, or a common work schedule with the vehicle operator, the set of similar vehicular trips having similar travel paths as the unlabeled vehicular trip; and

accepting the user classification upon determining that the user classification matches the machine classification predicted by using the trained classification machine learning model and the community classification.

Assignments (2)

Continuity (1)

References Cited (65)