Exploiting LSTM Neural Network Algorithm Potentiality for Early Identification of Delayed Graduation in Higher Education

Adoption of deep learning classification algorithms in the domain area of higher education provides exploratory predictive data analytics able to exploit students’ academic behavior. Concretely, student retention and success are critical concerns in higher education globally. Timely identification of potential delays in graduation is essential for universities to provide effective interventions and support, ensuring students’ progress efficiently and maintaining high graduation rates, thereby enhancing institutional reputation. This study examines data from a typical computer science department of a central Greek university, covering student performance for almost two decades (1999-2018). Through extended data preprocessing, we developed a robust dataset focusing on key courses indicative of students' likelihood to graduate on time or experience delays. We employed a deep learning Long Short-Term Memory (LSTM) Neural Network algorithm, leveraging this dataset to classify and predict students' final academic outcomes. Our findings reveal that early-semester performance data can successfully forecast graduation timelines, enabling proactive educational strategies to support student success during their studies at the university.