Human action recognition in sports is an important research area that contributes to the well-being and performance analysis of individuals. During the preceding years, the recognition of human movements and gestures in sports has become essential for evaluating, guiding, and improving athletic performance. It aids in identifying injuries and potential physical issues in the human body. In this paper a novel approach is presented based on two main phases; first phase which is features extraction and representation in which human pose estimation were detected by an algorithm called MediaPipe, once we obtained these poses, we contributed by representing these poses into graphs where nodes represent the coordinates of the poses and edges are the links between them. In the second steps we used Graph Neural Networks to classify these activities. The developed approach was tested on Olympics sports dataset and has shown exceptional performance in recognizing human activities, surpassing the current standards in human action recognition.