A robust iterative algorithm of weighted least squares support vector machine (WLS-SVM) was proposed to overcome the negative influence of outliers on spectral analysis. A novel method to calculate regression error was proposed to solve the iterative convergence problem in original WLS-SVM; to improve the robustness of original WLS-SVM, the formula for computing weighted value was also revised: the median value of regression error was selected as criteria. This algorithm has been applied to spectral quantitative analysis. Compared with original algorithm, the experimental results show the proposed algorithm is convergent and the breakdown point value is approximately 35%.

Key words: spectral analysis, least squares support vector machine, convergence, robustness, breakdown point