The field of student learning analytics is increasing significantly in educational settings, allowing improving academic outcomes and preventing dropouts. Artificial Intelligence (AI) plays a vital role in addressing challenges like predicting student performance, categorizing students, personalizing learning pathways, and identifying dropout risks using educational data. However, processing these dataset raises legal and ethical concerns. Federated Learning (FL) methodologies are emerging as a solution to these issues. They enable decentralized machine learning while preserving data privacy. This study introduces a novel Federated Learning approach tailored for student dropout prediction. This approach prioritizes academic centers with high-quality data by assigning weight-based importance during model aggregation. The proposed approach is evaluated starting from a free available dataset. The results demonstrate that the FL approach is robust and outperforms alternative approaches in predicting student dropouts.