Modeling Insurance Fraud Detection Using Imbalanced Data Classification

This paper proposes an innovative insurance fraud detection method to deal with the imbalanced data distribution. The idea is based on building insurance fraud detection models using Decision tree (DT), Support vector machine (SVM) and Artificial Neural Network (ANN), on data partitions derived from under-sampling (with-replacement and without-replacement) of the majority class and merging it with the minority class. Throughout the paper, ten-fold cross validation method of testing is used. Its originality lies in the use of several partitioning under-sampling approaches and choosing the best. Results from a publicly available automobile insurance fraud detection data set demonstrate that DT performs slightly better than other algorithms, so DT model was used to compare between different partitioning-under-sampling approaches. Empirical results illustrate that the proposed model gave better results.