Selection Algorithm For Rough Set Based On AIC Information Criterion

Nowadays,with the increasingly development of science technology,a large amount of data can be produced in every minute.These data are becoming more and more complex,and the scale is growing,forming a large number of complex and heterogeneous data.Data has become the main carrier of knowledge information in many fields that are closely related to human beings,and data mining is arises from the urgent need of human to obtain useful knowledge from the data.Rough set algorithm as one of the most commonly used data mining classification method,introduced by Pawlak,which classification rules are easy to understand,is not affected by data distribution control.It has been applied to many fields such as transportation,biological science,artificial intelligence and so on.The theoretical research of rough set is also in depth,which involves the improvement of rough set construction algorithm,the algorithm of the data problems such as uncertainty and fuzziness and so on.For the same data set,using different rough set construction algorithm,the classification rules are not always the same.How to choose a rough set of high prediction accuracy has important theoretical value and practical significance in the application.The existing rough set evaluation criteria are mainly the misclassification error.However,the lowest misclassification error of rough set does not show that it has the highest prediction accuracy.Especially when there is small difference of misclassification error among many rough sets,the ones chosen by misclassification criterion do not always have the highest prediction accuracy.In view of this,introduced rough set selection algorithm based on AIC,using rough set classification rules as the explained variable,the rough set is represented by a Logistic model,the AIC value of rough set is defined as the AIC value of the fitted model,that the AIC criterion is provided to choice rough sets.Rough set with small AIC value often has higher prediction accuracy than the one with large AIC value.The simulation results show that the selection algorithm for rough set based on AIC can choice out rough set with high prediction accuracy,its selection effect sometimes is higher than the selection algorithm for rough set based on misclassification error standard.Especially when there is small difference of misclassification error among many rough sets,new algorithm has bigger probability to choice out rough set with the highest prediction accuracy than misclassification criterion.