Nonlinear robust modeling base on least trimmed squares regression

Due to low breakdown point of existing nonlinear robust modeling algorithms, a novel robust modeling algorithm based on least trimmed squares is proposed. This algorithm is based on linear least trimmed squares regression. Confidence interval of normal distribution is used to select outliers, and least square support vector machine regression is applied for nonlinear modeling. Simulation results show the breakdown point for the algorithm can exceed 45%, and it is more sensitive in outlier detection than other nonlinear robust modeling algorithms.