PREDICTION AND PRIORITIZATION OF STUDENT DROPOUT VIA FEW-SHOT LEARNING: AN APPROACH TO DECISION SUPPORT

Dropout in higher education is a relevant challenge, especially in courses with a large number of students and limited resources for interventions. Early identification of at-risk students is essential to support institutional decision-making. This work proposes an approach based on few-shot learning, using Prototypical Networks, to predict evasion in scenarios with little data. The model is integrated with a prioritization strategy, allowing the construction of risk rankings to optimize resource allocation. The approach was evaluated with real data and compared to traditional machine learning models. The results show a significant gain in the recall metric, greater than 90%, compared to about 60% in conventional models, indicating a greater ability to identify at-risk students. Statistical tests confirm the significance of the results, and the solution features high scalability for efficient analysis of at-risk students, even with little data.