LOSS PREDICTION AND AGING CLAIM PREDICTION OF AUTO INSURANCE

An auto insurance business has grown dramatically because of the rise in voluntary
motor insurance policy. As a result, the number of claims increases every year. Also, the
insurance company is unable to determine the actual loss, when a claim is made, to use this
information to set a reserve fund for this claim payment. Furthermore, a claim age of each
claim may be different. This research aims to construct a model for estimating loss from car
accidents and to classify claim ages. The data of auto insurance claim from 2019 to 2022 of a
non-life insurance company are used in this study. Classification and regression (CART) for
decision tree method and K-nearest neighbor technique are used to construct models for
estimating losses from car accidents. After comparing model efficiencies by using a root mean
square error and a mean absolute error, it is revealed that the decision tree has more efficiency.
The insurance claims can be categorized into twenty-five groups according to the factors used
in this study. For a claim aging classification, a random forest model is used to investigate
factors affecting claim age. The affecting factors consisting of car damage cost, Days from
Policy Inception to Accident Date, sum insured and car age. These factors are used to classify
the claim age by using multinomial logistic regression into three groups according to the
distribution of data. The result of performance measurement shows that an accuracy of the
classification is 49.54 percents. Moreover, the data of each group are analyzed and determined
the period for adjusting claim estimation in three times as follows: The periods for adjusting
the claim reserve of the first group are 26 days, 32 days, and 38 days, respectively. The periods
of the second group are 5 2, 6 4, and 80 days, respectively, while those of the third group are
127, 164, and 242 days, respectively.