Wearable sensing devices can reliably track players' mobility, revolutionizing sports training. However, current sensing electronics face challenges due to their complex structures, battery dependence, and unreliable sensing signals. Here, a tennis training system is demonstrated using machine learning based on elastic self-powered sensing yarns. By employing a simple and effective strategy, piezoelectric nanofibers and triboelectric materials are integrated into a single yarn, enabling the simultaneous translation of both triboelectric and piezoelectric signals. Additionally, these yarns exhibit outstanding processability, allowing them to be machine-knitted into self-powered sensing fabrics. Due to their great sensitivity, these sensing yarns and fabrics may detect human movement with great precision. Machine learning algorithms can classify and interpret these signals to recognize various human motions. The developed tennis training system aims to maximize its benefits and provide comprehensive training for both players and coaches. This work enhances the applicability of self-powered sensing systems in smart sports monitoring and training, advancing the field of intelligent sports training.