Aiming at the situation that least squares support vector machine is sensitive to outliers and lacks robustmess, an improved leastsquares support vector machine algorithm considering sample outliers is proposed. The algorithm first ealeulates a LOF for each datasample using the loeal outlier factor detection algorithm, and can efectively divide the samples into normal and abnormal samplesaceording to their factor values, and then separately set sample weights for different samples. The algorithm effectively ensures that the weight of abnormal samples is reduced while the weight of normal samples is not affected, so that the least squares support vectormachine can achieve the optimization of the objective funetion while ensuring that the normal data information is not lost, so as to improvethe robustness of the model, Finally,"information entropy"and "average particle distance" are introduced to improve the particle swarmalgorithm, which is applied to the parameter optimization of the model. Experiment simmlation shows that the algorithm can effectivelyimprove the robustness of the model. With the increase of abmormal samples, the accuracy of the model is improved by about 67%.Keywords:improved least square support vector machines; loeal outlier factor deteetion algorithm; improved PSO algorithm