Evaluation of the performance of hydrokinetic turbines is an essential subject in the renewable energy field. In this regard, advanced machine learning methods including hybrid CatBoost and standalone Catboost and linear regression were used for the first time for the straight channel data and channel bend data. The hybrid models were developed using whale optimization (WO), grey wolf optimization (GWO), and Bayesian optimization (BO) algorithms. Sensitivity analysis was also conducted to find the most influential parameters in the best model. Results showed that the hybrid models can predict the goal parameters, including the coefficient of the power (CP), the maximum value of the coefficient of the power (CPmax), and the tip speed ratio corresponding to the Cpmax (TSRmax) better than the standalone models. Among the models, GWO-CB is the best model, even in the k-fold cross-validation scenario. The R2 value of the GWO-CB reached 0.97, 0.94, and 0.84 for the CP, CPmax, and TSRmax, respectively. Among the effective parameters, the lateral position of the returning blade deflector and the number of stages had the maximum and minimum effects on the best model, respectively. The effect of the position of the turbines in the channel bend, which was considered in this research for the first time, was significant in the models.