Applying machine learning to telematics data to predict accident outcomes
Systems and methods in accordance with embodiments of the invention can obtain and use a variety of telematics and other data to predict potential claim or accident outcomes arising from a variety of incidents, such as automobile accidents. The claim or accident outcomes, such as parts of the vehicle damaged, accident category, expected liabilities, and many others, can be predicted based on the telematics and other data, such as accelerometer data, heading data, location/GPS, barometer, gyroscope, magnetometer data, and the like. The machine learning classifiers that can be generated are trained on historical data, consisting of claims, telematics, and/or other relevant data. In a variety of embodiments, the telematics data can be captured using a telematics device installed in the vehicle and/or via a mobile device associated with the vehicle.
1 . A method, comprising:
obtaining, by a classification server system having a wireless network connection with a telematics device, telematics data comprising acceleration data, speed data, and heading data for a vehicle;
based on a determination that the telematics data indicates that an acceleration of the vehicle has exceeded a predefined g-force threshold in a predefined number of samples, increasing an upload frequency of the telematics data, wherein the telematics data indicating that the acceleration of the vehicle has exceeded the predefined g-force threshold in the predefined number of samples indicates that the vehicle has been involved in a vehicular accident, wherein the increased upload frequency of the telematics data causes an increased amount of the telematics data to be transmitted per each period of time of transmittal of the telematics data, wherein the telematics data includes data that occurred before the acceleration of the vehicle exceeded the predefined g-force threshold, data that occurred at the time the acceleration of the vehicle exceeded the predefined g-force threshold, and data that occurred after the acceleration of the vehicle exceeded the predefined g-force threshold;
obtaining, by the classification server system, the telematics data at the increased upload frequency;
generating, by the classification server system, common sampling rate data based on the telematics data, wherein the common sampling rate data comprises the acceleration data sampled at a first sampling rate using an accelerometer and location data sampled at a second sampling rate using a GPS receiver, the first sampling rate and the second sampling rate converted to a common sampling rate;
generating, by the classification server system and based on the common sampling rate data, a data set comprising a set of latent features associated with divided severity level buckets by:
training a machine-learning classifier to label the data set into a plurality of severity level buckets corresponding to at least a major severity label and a minor severity label by providing the machine-learning classifier:
historical accident outcome data including repair cost amount values; and
predefined cost amount thresholds which define the major severity label and the minor severity label;
obtaining, by the machine-learning classifier of the classification server system, accident predictions for the vehicle involved in the vehicular accident by classifying the data set based on historical telematics data and the historical accident outcome data, the historical telematics data and the historical accident outcome data having a predefined statistical similarity to the vehicular accident;
determining, by the classification server system, a decision for the vehicle based on the accident predictions;
based on the decision for the vehicle, automatically causing a tow truck to be dispatched to a location associated with the vehicle, with a target destination to which the tow truck is to tow the vehicle to;
calculating a predicted cost associated with the accident predictions by providing the machine-learning classifier the accident predictions for the vehicle; and
optimizing the machine-learning classifier by automatically updating a hyperparameter of the machine-learning classifier, based on the predicted cost associated with the accident predictions for the vehicle.
2 . The method of claim 1 , wherein the historical accident outcome data comprises data indicating parts damaged and liabilities for a vehicle involved in a historical vehicular accident.
3 . The method of claim 1 , wherein a classified data set comprises a set of damage areas for the vehicle and a probabilistic likelihood that each damage area of the vehicle is damaged based on the telematics data.
4 . The method of claim 3 , wherein the accident predictions further comprise a severity level for each damage area.
5 . The method of claim 1 , wherein the decision is selected from a group comprising of repairing the vehicle and totaling the vehicle.
6 . A computing device, comprising:
one or more processors; and
memory storing instructions that, when executed by the one or more processors, cause the computing device to:
obtain, using a wireless network connection with a telematics device, telematics data for a vehicle;
based on a determination that the telematics data indicates that an acceleration of the vehicle has exceeded a predefined g-force threshold in a predefined number of samples, increase an upload frequency of the telematics data, wherein the telematics data indicating that the acceleration of the vehicle has exceeded the predefined g-force threshold in the predefined number of samples indicates that the vehicle has been involved in a vehicular accident, wherein the increased upload frequency of the telematics data causes an increased amount of the telematics data to be transmitted per each period of time of transmittal of the telematics data, wherein the telematics data includes data that occurred before the acceleration of the vehicle exceeded the predefined g-force threshold, data that occurred at the time the acceleration of the vehicle exceeded the predefined g-force threshold, and data that occurred after the acceleration of the vehicle exceeded the predefined g-force threshold;
obtain, using the wireless network connection with the telematics device, the telematics data at the increased upload frequency;
generate common sampling rate data based on the telematics data, wherein the common sampling rate data comprises acceleration data sampled at a first sampling rate using an accelerometer and location data sampled at a second sampling rate using a GPS receiver, the second sampling rate is a same rate as the first sampling rate;
generate, based on the common sampling rate data, a data set comprising a set of latent features associated with divided severity level buckets by:
training a machine-learning classifier to classify the data set into a plurality of severity level buckets corresponding to at least a major severity label and a minor severity label by providing the machine-learning classifier:
historical accident outcome data including repair cost amount values; and
predefined cost amount thresholds which define the major severity label and the minor severity label;
obtain, by the machine-learning classifier, accident predictions for the vehicle involved in the vehicular accident by classifying the data set based on a set of historical outcome data for a historical vehicle accidents data set, the historical vehicle accidents data set having a predefined similarity to the vehicular accident;
determine a decision for the vehicle based on the accident predictions;
based on the decision for the vehicle, automatically cause a tow truck to be dispatched to a location associated with the vehicle, with a target destination to which the tow truck is to tow the vehicle to;
calculate a predicted cost associated with the accident predictions by providing the machine-learning classifier the accident predictions for the vehicle; and
optimize the machine-learning classifier by automatically updating a hyperparameter of the machine-learning classifier, based on the predicted cost associated with the accident predictions for the vehicle, thus enhancing prediction outcomes of the machine-learning classifier.
7 . The computing device of claim 6 , wherein the historical accident outcome data comprises data indicating parts damaged and liabilities for a vehicle involved in a historical vehicular accident.
8 . The computing device of claim 6 , wherein a classified data set comprises a set of damage areas for the vehicle and a probabilistic likelihood that each damage area of the vehicle is damaged based on the telematics data.
9 . The computing device of claim 8 , wherein the accident predictions further comprise a severity level for each damage area.
10 . The computing device of claim 6 , wherein the determined decision is selected from a group comprising of repairing the vehicle and totaling the vehicle.
11 . A non-transitory machine-readable medium storing instructions that, when executed by one or more processors, cause the one or more processors to perform steps comprising:
obtaining, using a wireless network connection with a telematics device, telematics data comprising acceleration data, speed data, and heading data for a vehicle;
based on a determination that the telematics data indicates that an acceleration of the vehicle has exceeded a predefined g-force threshold in a predefined number of samples, increasing an upload frequency of the telematics data, wherein the telematics data indicating that the acceleration of the vehicle has exceeded the predefined g-force threshold in the predefined number of samples indicates that the vehicle has been involved in a vehicular accident, wherein the increased upload frequency of the telematics data causes an increased amount of the telematics data to be transmitted per each period of time of transmittal of the telematics data, wherein the telematics data includes data that occurred before the acceleration of the vehicle exceeded the predefined g-force threshold, data that occurred at the time the acceleration of the vehicle exceeded the predefined g-force threshold, and data that occurred after the acceleration of the vehicle exceeded the predefined g-force threshold;
obtaining, using the wireless network connection with the telematics device, the telematics data at the increased upload frequency;
generating common sampling rate data based on the telematics data, wherein the common sampling rate data comprises acceleration data sampled at a first sampling rate using an accelerometer and location data sampled at a second sampling rate using a GPS receiver, the second sampling rate is a same rate as the first sampling rate;
generating, based on the common sampling rate data, a data set comprising a set of latent features associated with divided severity level buckets by:
training a machine-learning classifier to label the data set into a plurality of severity level buckets corresponding to at least a major severity label and a minor severity label by providing the machine-learning classifier:
historical accident outcome data including repair cost amount values; and
predefined cost amount thresholds which define the major severity label and the minor severity label;
obtaining by the machine-learning classifier, accident predictions for the vehicle involved in the vehicular accident by classifying the generated data set based on a set of historical outcome data for a historical vehicle accidents data set, the historical vehicle accidents data set having a predefined statistical similarity to the vehicular accident;
determining a decision for the vehicle based on the accident predictions;
based on the decision for the vehicle, automatically causing a tow truck to be dispatched to a location associated with the vehicle, with a target destination to which the tow truck is to tow the vehicle to;
calculating a predicted cost associated with the accident predictions by providing the machine-learning classifier the accident predictions for the vehicle; and
updating a hyperparameter of the machine-learning classifier, based on the predicted cost associated with the accident predictions for the vehicle.
12 . The non-transitory machine-readable medium of claim 11 , wherein a historical accident outcome data comprises data indicating parts damaged and liabilities for a vehicle involved in a historical vehicular accident.
13 . The non-transitory machine-readable medium of claim 11 , wherein a classified data set comprises a set of damage areas for the vehicle and a probabilistic likelihood that each damage area of the vehicle is damaged based on the telematics data.
14 . The non-transitory machine-readable medium of claim 13 , wherein the accident predictions further comprise a severity level for each damage area.
15 . The method of claim 1 , further comprising:
changing a transmit time for the telematics data based on the accident predictions for the vehicle.