Backpropagation-Based Prediction of On-Time Graduation in Informatics Engineering at Palangka Raya University

This paper presents a study on predicting the ontime graduation of Informatics Engineering students at Palangka Raya University using a neural network model optimized with the backpropagation algorithm. On-time graduation is a critical metric for both students and educational institutions, influencing academic success, institutional reputation, and accreditation standards. However, the Informatics Department at Palangka Raya University faces challenges with low on-time graduation rates, with a significant imbalance between students who graduate on time and those who do not. To address this issue, we developed a predictive model that incorporates both academic and non-academic factors. The dataset was processed using the SMOTE technique to handle class imbalance and was then split using a stratified method to ensure proportional representation in training and testing sets. The model, configured with two hidden layers and the Tanh activation function, demonstrated high predictive accuracy, achieving a balanced accuracy of $94 \%$ and an AUCROC score of 0.94. These results indicate the model’s robust ability to distinguish between students likely to graduate on time and those at risk of delayed graduation. Furthermore, the study identified GPA as the most influential factor in predicting ontime graduation, along with other academic and non-academic parameters. The findings suggest that the backpropagation algorithm is highly effective in educational data mining, providing valuable insights that can guide interventions to improve student outcomes. The study concludes by proposing the development of a user-friendly interface for the deployment of this predictive model within the university’s academic management system.