It is essential to have an accurate prediction of students' future performance in order to properly carry out the necessary pedagogical interventions that are required to assure students will graduate on time and with an acceptable degree. Even though there is a wealth of research on the topic of predicting student performance when it comes to finding solutions to problems or preparing for classes by utilizing data-driven methods, the topic of predicting student performance when it comes to completing degrees (for example, college programs) is much less researched and faces new challenges, there is a large amount of diversity among students in terms of their prior experiences and the courses they choose to take. The students' developing progress should be factored into the prediction. In this work, we offer a unique machine learning technique for forecasting student success in degree programs. This method may handle these important issues, and it is one of the main contributions of this research. The suggested technique is distinguished primarily by its two components. To begin, a structure with two layers, a bi-layered structure, is constructed for the purpose of creating predictions based on the changing performance states of students. Then, a strategy that is driven by data and is based on latent component models and ensemble progressive prediction (EPP) based matrix factorization is suggested for the purpose of determining the relevance of the course, which is essential for the construction of effective base predictors. We demonstrate that the suggested strategy achieves better performance than benchmark methods by conducting extensive simulations using a dataset of UCLA undergraduate student data that was gathered over the course of three years.