State Farm Mutual Automobile Insurance Company

Meghan Sims Goldfarb

Scott T Christensen

Gregory L Hayward

Nicholas U Christopulos

Taylor Michael Thiel

Erik Donahue

Anuj Dahiya

Techniques for implementing machine learning to improve claim handling are disclosed. In some scenarios, the machine-learning, analytics model may be trained in accordance with data that is relevant to insurance products, such as life and health insurance. A set of labeled historical claims each corresponding to a settlement amount may be analyzed to train an artificial neural network. A claim may be received from a user mobile device, and may be analyzed using the trained artificial neural network to predict a claim settlement, which may be used to generate a settlement offer. The settlement offer may be transmitted to the user's mobile device, and if a manifestation of acceptance is received from the user, then the claim may be automatically paid.

Implementing machine learning for life and health insurance claims handling

A method of determining damage to property includes inputting historical data into a machine learning model to identify an insured type, features, and/or characteristics. The method may include identifying a peril, repair and/or replacement cost of the vehicle by analyzing a digital image from a device of an insured, the digital image depicting damage to the vehicle. The method may include inputting the digital image into the trained machine learning model to identify a type, feature, and/or characteristic of the vehicle, and may include identifying a peril, repair, and/or replacement cost associated with the vehicle. A method may include receiving and/or retrieving free-form text associated with an insurance claim and/or a vehicle, identifying at least one key word composing the free-form text, and determining based on the at least one key word a cause of loss and/or peril that caused damage to the vehicle.

Automobile monitoring systems and methods for detecting damage and other conditions

Machine learning systems, methods, and techniques for detecting damage and/or other conditions associated with a building, land, structure, or other real property using a real property monitoring system are disclosed. The property monitoring system is used in conjunction with machine learning techniques to determine and/or predict various conditions associated with the real property, including particular damage thereto, e.g., based upon dynamic characteristic data obtained via on-site sensors, static characteristic data, third-party input descriptive of an event impacting the building, etc. Accordingly, damage and/or loss associated with the building/real property is more quickly and/or accurately ascertained so that suitable mitigation techniques may be applied. In some scenarios, previously undetectable or uncharacterized damage and/or other conditions may be discovered and mitigated.

Real property monitoring systems and methods for detecting damage and other conditions

A method for attributing vehicle telematics data to individuals may include receiving vehicle telematics data collected by a data collection device during a plurality of trips. Subsets of the vehicle telematics data may correspond to different trips, and may be used to generate respective metric sets. Each metric set may include metrics indicative of different driving behaviors and/or different features of a driving environment. The method may also include retrieving, from a policy database, policy information pertaining to an insurance policy associated with the data collection device, and determining, based upon the policy information, a number of disclosed drivers associated with the insurance policy. A statistical analysis that includes executing a clustering algorithm may be performed on the metric sets, and, based upon the results, at least some of the metrics and/or at least some of the subsets of vehicle telematics data may be assigned to the disclosed drivers.

Driver identification for trips associated with anonymous vehicle telematics data

Historical vehicle telematics data, corresponding to trips known to have been driven by specific persons within a limited pool of potential drivers, may be processed to generate a statistical model that is customized for those persons. Once generated, the custom statistical model may be used to process vehicle telematics data from trips where it is known that the driver was one of the drivers in the pool of drivers, but the specific identify of the driver is not known. Because the statistical model is specifically optimized or designed to distinguish among the drivers in the pool, the model may be more accurate than universal driver identification models.

Driver identification among a limited pool of potential drivers

A method for attributing vehicle telematics data to individuals may include receiving vehicle telematics data collected by a data collection device during a plurality of trips. Subsets of the vehicle telematics data may correspond to different trips, and may be used to generate respective metric sets. Each metric set may include one or more metrics each indicative of a different driving behavior or a different feature of a driving environment. The method may also include retrieving, from a policy database, policy information pertaining to an insurance policy associated with the data collection device, and determining, based upon the policy information, a number of disclosed drivers associated with the insurance policy. A statistical analysis may be performed on the metric sets, and, based upon the results, at least some of the metrics and/or at least some of the subsets of vehicle telematics data may be assigned to the disclosed drivers.

Growing community of inventors

Bloomington, IL, United States of America

Meghan Sims Goldfarb