| CPC G06Q 10/1091 (2013.01) [G06N 20/00 (2019.01)] | 20 Claims |
|
1. A computer-implemented method for classifying a vehicular trip, the method comprising:
retrieving a first set of historic device interaction data stored in a memory device of an electronic device associated with a first set of historic vehicular trips during which vehicle operators operated vehicles for work;
retrieving a second set of historic device interaction data stored in the memory device of the electronic device associated with a second set of historic vehicular trips during which the vehicle operators operated the vehicles for personal use;
training a classification model based on a training dataset to generate a trained classification model that is trained to predict a probability that a vehicular trip is used as a work trip, wherein the training dataset corresponds to a plurality of vehicular trips, wherein the training dataset comprises the first set of historic device interaction data and the second set of historic device interaction data for the vehicle operators and the vehicles operated by the vehicle operators, and wherein the classification model is trained using hyperparameter tuning, cross-validation, and Bayesian methods to optimize performance of the trained classification model across diverse datasets, such that:
after receiving a first input of device interaction data selected from the first set of historic device interaction data, the trained classification model classifies a first associated vehicular trip as work based on at least a threshold of trip-level probabilities; and
after receiving a second input of device interaction data selected from the second set of historic device interaction data, the trained classification model classifies a second associated vehicular trip as personal use based on at least the threshold of trip-level probabilities;
tuning, using group cross-validation and hyperparameter optimization techniques comprising Bayesian methods, the trained classification model to further customize the trained classification model at an individual level corresponding to each of the vehicle operators based on at least the threshold of trip-level probabilities;
retrieving (i) a set of unlabeled device interaction data, (ii) a set of unlabeled path conditions, and (iii) a set of operation behaviors, each of which is stored in the memory device of the electronic device and is associated with an unlabeled vehicular trip during which a vehicle operator operated an unlabeled vehicle; and
classifying, using the trained classification model, 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 device interaction data, a first set of baseline device interaction features and first operation behaviors associated with the first set of historic vehicular trips;
identifying, based at least in part upon the second set of historic device interaction data, a second set of baseline device interaction features and second operation behaviors associated with the second set of historic vehicular trips;
identifying, based at least in part upon the set of unlabeled device interaction data and the set of operation behaviors, a set of representative device interaction features and representative operation behaviors associated with the unlabeled vehicular trip, wherein the set of representative device interaction features are calibrated based at least in part upon the set of unlabeled path conditions, and wherein the representative operation behaviors associated with the unlabeled vehicular trip comprise at least one of acceleration, braking, cornering, lane changes, speed compared to speed limit, steering, magnitude of jerk, or magnitude of swerve;
comparing the set of representative device interaction features and the representative operation behaviors against the first set of baseline device interaction features and the first operation behaviors, and against the second set of baseline device interaction features and the second operation behaviors; and
classifying the unlabeled vehicular trip by at least one of:
after determining that the set of representative device interaction features and the representative operation behaviors deviate from the first set of baseline device interaction features and the first operation behaviors less than or equal to a first deviation threshold, classifying the unlabeled vehicular trip as work; or
after determining that the set of representative device interaction features and the representative operation behaviors deviate from the second set of baseline device interaction features and the second operation behaviors less than or equal to the first deviation threshold, classifying the unlabeled vehicular trip as personal use.
|