Abstract: A machine learning algorithm——Support Vector Machines (SVM) was introduced into the field of soft rock engineering supporting design. An improved Support Vector Machines Regression (SVR) algorithm was presented to meet the needs of this problem and the corresponding calculation code was programmed, It is concluded that a high degree of prediction accuracy and a very good generalization can be obtained with small quantity of learning samples using this algorithm from the calculated results of an engineering instance. It can avoid the overfitting problem of artificial neural network (ANN) which brings the difficulty in determining the parameters of ANN. It facilitates users to a great extent and provides a new way in the supporting design of similar engineering.