Primary lymphoma of bone (PLB) significantly worsens in prognosis with distant organ involvement, leading to decreased survival rates. Early detection and appropriate intervention are critical, yet systematic treatment strategies and predictive models are lacking. This study aims to develop and validate a machine learning model to predict the risk of distant metastasis in PLB and identify relevant risk factors. Utilizing the SEER database from the National Institutes of Health, 690 PLB patients diagnosed between 2000 and 2021 were analyzed to construct machine learning models. The models’ performance was evaluated using ROC AUC, with the best-performing model being further validated on an external cohort of 142 PLB patients from Changzheng Hospital, demonstrating model generalizability. SHAP values were used to visualize disease-related risk factors. A web-based calculator employing the optimal model was developed to predict PLB distant organ involvement risk. In total, 832 patients were included, with 666 experiencing distant metastasis. The Random Forest model showed the best predictive capability, achieving an internal accuracy of 0.852 and AUC of 0.907. External validation confirmed its performance, with an accuracy of 0.929 and AUC of 0.977.This study presents an RF algorithm-based model to assist clinicians in making informed clinical predictions for PLB patients.