As for the bladed disk structures, aiming to predict the occurrence probability of events featuring extremely high vibration response levels induced by mistuning, a probability prediction approach for extreme response events (EREs) based on the importance sampling method is proposed. A binary indicator function is employed to characterize the EREs of mistuned bladed disks. The importance sampling density function is constructed using the Gaussian mixture model, and the update rules for the parameters of the Gaussian mixture model are derived based on cross-entropy minimization. Moreover, an improved JAYA algorithm, in conjunction with cluster analysis, is utilized to furnish the initial values for the update of the Gaussian mixture model, which expedites the parameter update process. Then, the probability of EREs is predicted via importance sampling. Finally, by applying the proposed probability prediction method, the occurrence probability of EREs of mistuned bladed disks is calculated based on the lumped-parameter model of the bladed disk. The computational results are compared with those obtained from the direct Monte Carlo (DMC) method. This comparison validates the effectiveness of the proposed method. Moreover, it demonstrates that the proposed method exhibits higher computational efficiency and enables the acquisition of a narrower confidence interval.