<p>The tribological performance of self-lubricating composites is a complex system influenced by material properties and test conditions. Data-driven approaches, including machine learning (ML) algorithms, can provide a more comprehensive understanding of complex problems influenced by multiple parameters. The correlation between the tribological wear performance of graphite-filled polytetrafluoroethylene (PTFE) composites and the inherent material properties and test variables was investigated using two-parameter relationship analyses methods. Then a generative adversarial network (GAN) data augmentation method is proposed to address the problem of poor model generalization due to data scarcity, and five machine learning algorithms (K-nearest neighbor, eXtreme gradient boosting, gradient boosting decision tree, random forest regression and gradient boosting regression) are used to predict tribological properties based on the augmented data. The results show that the GBR model has good prediction ability for friction coefficient and wear rate, with R² values of 0.9476 and 0.9346, respectively. The feature significance analysis shows that normal load and sliding speed have the greatest influence on the friction coefficient, and graphite content and matrix hardness have the greatest influence on the wear rate. Finally, the validity of the model was experimentally verified. This study provides a new method for predicting the tribological properties of composites using machine learning techniques and provides a theoretical basis for material design and optimization.</p>