A comparative study of data mining algorithms in the prediction of auto insurance claims

Abstract

Insurance claims are a significant and costly problem for insurance companies. The prediction of auto insurance claims has been a challenging research problem for many auto insurance companies. Identifying the risk factors which are affected for the high number of claims and denying them may lead to increased corporate profitability and keep insurance premiums at a below rate. The key objective of conducting this study is to examine the data mining techniques in developing a predictive model in support of auto insurance claim prediction and a comparative study of them. The research was carried out by using Artificial Neural Network (ANN), Decision Tree (DT) and Multinomial Logistic Regression (MLR) to develop the prediction model. The results indicated that the ANN is the best predictor with 61.71% overall classifier accuracy. Decision tree came out to be the second with 57.05% accuracy and the logistic regression model indicated 52.39% accuracy. Parameters of optimal NN model gives 6 input neurons and 7 minimum hidden neurons with 0.15 learning rate. The comparative study of multiple prediction models provided us with an insight into the relative prediction ability of different data mining methods. The comparison of the results of the decision tree and neural network models showed an interesting pattern. Policies that are misclassified by one model are correctly classified by the other. This might be an indication that the combination of the models could result in a better classification performance.