Efficient Evolutionary Data Mining Algorithms Applied to the Insurance Fraud Prediction

This study proposes two kinds of Evolutionary Data Mining (EvoDM) algorithms to the insurance fraud prediction. One is GA-Kmeans by combining K-means algorithm with genetic algorithm (GA). The other is MPSO-Kmeans by combining K-means algorithm with Momentum-type Particle Swarm Optimization (MPSO). The dataset used in this study is composed of 6 attributes with 5000 instances for car insurance claim. These 5000 instances are divided into 4000 training data and 1000 test data. Two different initial cluster centers for each attribute are set by means of (a) selecting the centers randomly from the training set and (b) averaging all data of training set, respectively. Thereafter, the proposed GA-Kmeans and MPSO-Kmeans are employed to determine the optimal weights and final cluster centers for attributes, and the accuracy of prediction for test set is computed based on the optimal weights and final cluster centers. Results show that the presented two EvoDM algorithms significantly enhance the accuracy of insurance fraud prediction when compared the results to that of pure K-means algorithm.