Xufre, PatríciaNunes, Luís CatelaPereira, Alexandra Costa2024-10-302024-10-302024-01-102024-01-10http://hdl.handle.net/10362/174304Effectively predicting student failure is crucial for timely resource allocation and preventing academic difficulties. This thesis explores conditions where machine learning model predictions should be trusted over teachers' judgments, recognizing potential disparities. Beginning with model building, one investigates the impact of both teachers' and students' features on the model-teacher interplay. Contrary to expectations, teachers' features do not significantly influence this dynamic, but a closer analysis of students' reveals the significance of psychological factors and academic achievement levels. The model exceeds for high achieving students and those with more distinctive psychological patterns. This research provides relevant insights into shaping reliable educational predictions in diverse academic scenarios.engMachine learning in educationPredictive modelingEconomics of educationInterplay teacher-modelmaster thesis203600584