Dynamic resource prediction in cloud computing can provide support for allocating resources required by complex system simulation (CSS) on demand, which can improve the simulation performance and resource utilization. However, the resource requirements have strong volatility because of the dynamic changes of simulation entities in the CSS applications, and few limited dynamic prediction models can predict the resource with strong volatility. In this study, a probabilistic approach using stacking ensemble learning, which integrates random forest, long short-term memory networks, linear regression, and Gaussian process regression, is proposed to predict the cloud resources required by the CSS applications. The proposed approach can quantify the uncertainty information in the cloud resource prediction. Experiments show that the proposed probabilistic approach using stacking ensemble learning can achieve better performance compared with other resource prediction approaches.