Paragangliomas and pheochromocytomas (PPGLs) are rare neuroendocrine tumors whose metastatic behavior remains difficult to predict using conventional clinical markers. Recent advances in machine learning (ML) offer new opportunities for improving diagnostic support, but their adoption in healthcare is often limited by the opacity of model decision-making. In this work, we present a machine learning framework designed to predict metastatic potential in PPGLs, integrating multiple classification algorithms with explainable artificial intelligence (XAI) techniques. The proposed approach emphasizes model interpretability, combining data preprocessing strategies, classifier evaluation, and SHAP-based analysis to uncover the most influential features driving predictions. Our results confirm that ensemble-based models, particularly Random Forest, provide high accuracy and interpretable outputs, which are essential for clinical acceptance. This study highlights the importance of transparent ML models in supporting oncological decision-making and outlines future directions for enhancing trust and usability in medical applications.