Accurate predictions of students' future performance are crucial for implementing effective pedagogical interventions that ensure timely graduation and satisfactory academic outcomes. While there is extensive research on predicting student performance for specific tasks and course preparations using data-driven methods, there is significantly less focus on predicting overall degree completion. This gap presents unique challenges, especially given the diverse backgrounds and course selections of students. Moreover, students’ evolving progress must be incorporated into these predictions. In this work, we introduce an innovative machine learning approach to forecast student success in degree programs. Our method addresses these critical issues and represents a significant contribution to the field. It is characterized by two main components. First, we propose a bi-layered structure designed to generate predictions based on the dynamic performance states of students. Second, we present a data-driven strategy that utilizes latent component models and ensemble progressive prediction (EPP) based matrix factorization to assess course relevance, which is vital for developing effective base predictors.

Student performance, Ensemble progressive prediction, machine learning.