ANALISA PRESTASI MAHASISWA MENGGUNAKANJARINGAN SARAF TIRUAN METODE BACKPROPAGATION

Abstract

Students’ academic performance can serve as an indicator to support the monitoring of learning processes and the provision of early academic assistance. This study applies an Artificial Neural Network (ANN) using the Backpropagation algorithm to predict students’ cumulative Grade Point Average (GPA) at the end of the fourth semester based on Semester 1 to Semester 3 GPAs, the percentage of credits successfully completed, and the number of repeated courses. The data were processed through selection, anonymization, and cleaning, followed by partitioning into 70% for training, 15% for validation, and 15% for testing. The data were also normalized to a range of 0.1–0.9 using parameters calculated exclusively from the training set. Nine combinations of hidden-layer architectures and learning rates were evaluated, with each configuration tested five times using different random seeds. Since the available research files did not contain actual academic records, the quantitative results reported in this study were obtained using 320 anonymized/simulated test records to verify the experimental procedure rather than to establish empirical findings for Universitas Budi Darma. The 5-5-1 architecture with a learning rate of 0.10 achieved the lowest average validation RMSE of 0.119 ± 0.009. On the test dataset, the selected model achieved an MSE of 0.0162, an RMSE of 0.1274, an MAE of 0.1035, and an R² of 0.9415. The verification results indicate that Backpropagation can model nonlinear relationships among academic indicators. However, operational implementation requires further validation using official academic records.