Abstract: To reduce the influence of the gear wear data that contains noise on the robustness of least squares support vector machine (LSSVM) model, the data was modeled and forecasted by iteratively robust least squares support vector machine (IRLSSVM) . First, model process robustness was assured by increasing weight function iteration times; Second, the IRLSSVM hyper-parameter was optimized based on the method combined global optimization method CSA with local optimum method SM; Third, the robust cross validation was used as CSA- SM algorithm objective function to improve IRLSSVM model robustness of parameter optimization process; Finally, numerical experiment was carried out by using K727840 ZW gearbox data. result shows that the proposed method is effective.