In view of the shortcomings of current TBM data mining capability and tunnelling parameters optimization of TBM performance prediction as well as the unmanned driving in future, this paper examines the feasibility of least squares support vector machine (LSSVM) technology in the parameter prediction of TBM tunnelling. Four important parameters including the cutter torque, cutter thrust, total propulsion and propulsion speed of the ascending section were extracted from the TBM excavation data of the Yinsong Diversion Project to model, and this study predicted the mean value of the stable section. The influence on model prediction performance was discussed from the aspects of model size and parameter selection. The numerical results show that the LSSVM model established by uniformly extracted samples from the original data, RBF kernel function and 10fold cross validation can predict above four parameters in the stable segment more accurately. Therefore, the LSSVM machine learning method is a scientific and feasible method to predict TBM tunnelling parameters.