Cloud datacenter operational time series workloads tend to exhibit nonlinear patterns as a result of heterogeneous and auto-scalable infrastructure for resource provisioning, and high fluctuating user workloads. Many statistical and machine learning models are implemented on cloud workloads without any prior nonlinearity tests, which either underestimate the evaluated models or unnecessarily lead to the assemble of multiple models. Accordingly, the Identification of such nonlinear patterns is essential for the selection of the right models that enhance forecasting performance, anomaly discovery, and on-demand resource provisioning. To detect nonlinear behavior, we implemented various statistical tests, the BDS (Brock-Dechert-Scheinkman) test, the Tsay test, and Lyapunov Exponent (LE) Test on real-time GWA-BitBrain cloud performance metrics workload. Each method’s performance is measured by sensitivity to nonlinearity, noise, and autocorrelation robustness, and competence for different temporal structures. Additionally, a nonlinear machine learning model was implemented to confirm the nonlinearity of the dataset. The results provide useful suggestions for confirming the presence of nonlinearity in cloud workload and a guide for recommending the nonlinear type machine learning for precise resource prediction and offering QoS-based cloud infrastructure.