Abstract: Aimed at the problem that twin support vector machine (TWSVM) has poor performance in dealing with outlier classification, a generalized robust distance measure classification method based on Laplacian kernel correlation entropy is proposed. A bounded adaptive L_θε loss function was proposed. In the learning process, different loss functions could be selected by adaptive parameters θ. A correlation entropy induced robust distance measure based on Laplace kernel was proposed, and its boundedness, non-convexity, smoothness and approximation were proved. An adaptive robust twin support vector machine learning framework was introduced. Experimental results on several data sets show that the proposed method is robust to feature noise and outliers.