Prediksi Kelulusan Mata Kuliah Mahasiswa Berdasarkan Data Histori Pertemuan Perkuliahan Dengan Pendekatan Deep Learning

Setiadi, Keyisa Raihan Illah (2025) Prediksi Kelulusan Mata Kuliah Mahasiswa Berdasarkan Data Histori Pertemuan Perkuliahan Dengan Pendekatan Deep Learning. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025211002-Keyisa-Raihan-BukuTA.pdf - Accepted Version Restricted to Repository staff only Download (8MB) | Request a copy

Abstract

Tingkat ketidaklulusan mahasiswa yang masih tinggi menuntut adanya sistem prediksi yang akurat untuk membantu institusi pendidikan melakukan intervensi dini. Penelitian ini bertujuan membangun model prediksi kelulusan mahasiswa berdasarkan data akademik menggunakan pendekatan deep learning. Prediksi dilakukan terhadap kelulusan mahasiswa dalam suatu mata kuliah selama satu semester, dengan kelas target berupa biner: lulus dan tidak lulus. Data sintetis dihasilkan menggunakan empat algoritma, yaitu TVAE, CTGAN, CTABGAN, dan CTABGAN+. Hasil pengujian menunjukkan bahwa CTABGAN+ menghasilkan data sintetis dengan kualitas prediktif terbaik. Selanjutnya, proses preprocessing dilakukan melalui data cleaning, feature engineering, dan mix encoding yang terbukti memberikan kinerja model optimal. Evaluasi arsitektur deep learning menunjukkan bahwa model DNN dua layer dan CNN satu layer menghasilkan performa terbaik dibandingkan model berbasis RNN. Tahap hyperparameter tuning lebih lanjut menghasilkan konfigurasi optimal pada model CNN dengan optimizer RMSProp, learning rate 0,0001, dan batch size 16, yang mencapai accuracy 88,08%, precision 85,48%, recall 91,74%, dan F1-score 88,50%.
=================================================================================================================================
The high rate of student non-graduation requires an accurate prediction system to help educational institutions make early interventions. This research aims to build a student graduation prediction model based on academic data using a deep learning approach. Prediction is done on student graduation in a course for one semester, with the target class being binary: pass and fail. Synthetic data is generated using four algorithms, namely TVAE, CTGAN, CTABGAN, and CTABGAN+. Test results show that CTABGAN+ produces synthetic data with the best predictive quality. Furthermore, preprocessing is done through data cleaning, feature engineering, and mix encoding which is proven to provide optimal model performance. Evaluation of the deep learning architecture shows that the two-layer DNN and one-layer CNN models produce the best performance compared to RNN-based models. Further hyperparameter tuning stage resulted in the optimal configuration of CNN model with RMSProp optimizer, learning rate 0,0001, and batch size 16, which achieved accuracy 88.08%, precision 85.48%, recall 91.74%, and F1-score 88.50%.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Data Akademik, Data Sintetis, Deep Learning, Academics Data, Deep Learning, Synthetic Data
Subjects:	T Technology > T Technology (General) > T58.62 Decision support systems
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering
Depositing User:	Keyisa Raihan Illah Setiadi
Date Deposited:	24 Jul 2025 05:30
Last Modified:	24 Jul 2025 05:30
URI:	http://repository.its.ac.id/id/eprint/121130

Actions (login required)

View Item