The IEC standard 61400 − 12 − 1 recommends a reliable and repeatable methodology called ‘binning’ for accurate computation of wind turbine power curves that recognise only the mean wind speed at hub height and the air density as relevant input parameters. However, several literature studies have suggested that power production from a wind turbine also depends significantly on several operational variables (such as rotor speed and blade pitch angle) and incorporating these could improve overall accuracy and fault detection capabilities. In this study, a Gaussian Process (GP), a machine learning, data-driven approach, based power curve models that incorporates these operational variables are proposed in order to analyse these variables impact on GP models accuracy as well as uncertainty. This study is significant as it find out key variable that can improve GP based condition monitoring activities (e.g., early failure detection) without additional complexity and computational costs and thus, helps in maintenance decision making process. Historical 10-minute average supervisory control and data acquisition (SCADA) datasets obtained from variable pitch regulated wind turbines, are used to train and validate the proposed research effectiveness. The results suggest that incorporating operational variables can improve the GP model accuracy and reduce uncertainty significantly in predicting a power curve. Furthermore, a comparative study shows that the impact of rotor speed on improving GP model accuracy is significant as compared to the blade pitch angle. Performance error metrics and uncertainty calculations are successfully applied to confirm all these conclusions.