Abstract: OBJECTIVE To systematically evaluate machine learning (ML)-based prediction models for readmission risk in patients after total joint arthroplasty (TJA), and to provide an evidence-based foundation for future model development and clinical application. METHODS A systematic search was performed in PubMed, Cochrane Library, Embase, Web of Science, CNKI, VIP, and Wanfang databases from their inception to December 1, 2025. Data extraction and quality assessment were conducted according to the Checklist for Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modelling Studies (CHARMS). A systematic review and meta-analysis were then performed. RESULTS Fifteen retrospective studies were included. Meta-analysis results showed that the pooled area under the receiver operating characteristic curve (AUC) was 0.78 95% CI (0.71-0.83), P<0.001. Subgroup analysis indicated that advanced machine learning algorithms outperformed traditional logistic regression models. After sensitivity analysis by excluding two outlier studies, the pooled AUC was 0.79 95% CI (0.72-0.85), P<0.001, suggesting high robustness of the pooled effect size. CONCLUSIONS Machine learning algorithms demonstrate favorable predictive performance for assessing readmission risk following TJA. However, the overall quality of existing studies requires further improvement. Future research should leverage multi-center clinical data from China, standardize research designs, and emphasize external validation to develop localized risk prediction models.