International Journal of Computer Science Issues

More than a traditional journal...

The classical Fisher linear classification analysis is a well known linear classification procedure. This technique tends to minimize the misclassification error if the data set comes from a multivariate normal distribution. On the other hand, if the data set is contaminated, the misclassification error of this approach tends to increase. Relying on the classification performance of the classical Fishers technique for contaminated data set, robust Fisher linear classification analysis based on the minimum covariance determinant estimators was proposed. The objective of this procedure is to obtain maximum classification rate when the data set contain influential observations. This procedure only depends on the number of sample observations selected by the half set. The performance of the robust Fishers procedure strictly depends on the half set. Considering the classification performance of the classical and robust Fishers techniques, a robust M linear classification rule that utilizes all the entire data set is proposed to compare the classification performance of the above methods.