The data-driven fault detection techniques particularly artificial intelligent ones have many advantages over model-based methods is that not much information about system parameters is needed. In this work, a data-driven method based on machine learning random forest technique was proposed to instantaneous detecting and diagnosing a partial shading fault in a grid-connected PV system in real-time, a PV system installed in the desert area of Adrar, Algeria was used as a case study. The feasibility of the tree-based ensemble method (random forest) in detecting and diagnosing a partial shading fault in a grid-connected PV system was assured with high performance, the error was recorded with less than 1.3%.