There is endless buzz about the potential for machine learning in insurance, but there is now academic proof that AI can improve pricing for insurance companies.
Academics at Stanford University and the University of Warsaw successfully trained an algorithm to use Google Street View images to estimate the probability of loss for motor policies.
Academic researchers took historical loss data from a Polish insurance company and looked up policyholders' addresses on Google Street View. The researchers then used computer vision to extract information about the house - the type of dwelling, build quality, age etc. A machine learning algorithm was used to find correlations between the images of a policyholder's house and the probability they would have a motor claim.
Once applied to a new set of data, the additional factors provided by Street View images improved the predictive power of the model by 2%. This is a significant improvement given the previous model (using data on driver age, loss history etc) was only 8% better than a "null model" where all policies are priced the same.
Machine learning provides opportunities to integrate vast quantities of public data in pricing. These early experiments suggest the benefits to loss ratios could be dramatic.
The researchers’ method is straightforward. They began with a data set of 20,000 records of people who had taken out car insurance in Poland between 2013 and 2015. These were randomly selected from the database of an undisclosed insurance company. The results are something of a surprise. It turns out that a policyholder’s residence is a surprisingly good predictor of the likelihood that he or she will make a motor claim. “We found that features visible on a picture of a house can be predictive of car accident risk, independently from classically used variables such as age or zip code,” say Kidziński and Kita-Wojciechowska.
unknownx500
