Abstract: To predict rock burst tendencies in deep underground engineering, a prediction model of rock burst tendencies combining the principal component analysis (PCA) and improved GWO-SVM algorithm was proposed. Based on the formation mechanism of rock burst, the maximum tangential stress of rock σ_θ, the uniaxial compressive strength of rock σ_c, the uniaxial tensile strength of rock σ_t, the rock stress coefficient σ_θ/σ_c, the rock brittleness coefficient σ_c/σ_t, and the elastic energy index W_et were selected as estimate indexes for rock burst tendencies, and the three estimate indexes including F₁, F₂ and F₃ were obtained to optimize estimation index structure by PCA,which reduced the complexity of calculations. The original data set of rock burst was constructed by collecting 64 sets of rock burst cases at home and abroad, and the original rock burst data set and the pre-processed rock burst data set by PCA were used as input vectors of four machine learning models including the PNN model, the Elman model, the SVM model and the GWO-SVM model, and the prediction performances of each model were compared in different combinations of inputs in terms of prediction accuracy and running times. The results indicated that the overall performances of models were significantly improved after optimization by PCA. Comparison with those models before optimization by PCA, the calculation accuracy of improved models was enhanced by 6.25% to 12.5%, running times was shortened by 11.20% to 58.42%, and the prediction accuracy of the GWO-SVM model combined with PCA was up to 93.75%.