A weighted least squares support vector machine based on covariance matrix

In this paper, a weighted least squares support vector machine based on covariance matrix (CWLSSVM) is proposed. The structural information is vital for designing a good classifier in real-world problem, so the proposed method adds the covariance matrix of data to the objective function to identify the structure information in data. The LSSVM is sensitive to outliers, so a new weighting method is proposed according to distances between different types of samples and the center of samples. Different weights are assigned to different training samples in the error term of the objective function. Experimental results show that the CWLSSVM outperforms the LSSVM, the SVM and the ESVM.