Using Multi-class AdaBoost Tree for Prediction Frequency of Auto Insurance

In this paper, AdaBoost algorithm, a popular and effective prediction method, is applied to predict the prediction of claim frequency of auto insurance, which plays an important part of property insurance companies. Using a real dataset of car insurance, we reduce the frequency prediction problem to be a multi-class problem, in turn we employ the mixed method called multi-class AdaBoost tree (a combination of decision tree with adaptive boosting) as our predictor. By comparing its results with some most popular predictors such as generalized linear models, neural networks, and SVM, we demonstrate that the AdaBoost predictor is more comparable in terms of both prediction ability and interpretability. The later objective is particularly important in business environments. As a result, we arrive at the conclusion that AdaBoost algorithm could be employed as a robust method to predict auto insurance. It is important to practical contribution for insurance company in terms of conclusion explanation and decision making suggestions.