Implementasi Model Hybrid untuk Prediksi Kelulusan Mahasiswa Program Magister

Winarso, Raihan Adam Handoyo (2026) Implementasi Model Hybrid untuk Prediksi Kelulusan Mahasiswa Program Magister. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6025241043-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Prediksi kelulusan tepat waktu mahasiswa Pascasarjana (S2) menghadapi tantangan karena durasi studi yang singkat dan ketersediaan data historis yang terbatas. Model Machine Learning (ML) dan Deep Learning (DL) telah digunakan secara terpisah, namun masing-masing memiliki kelemahan. Model ML kurang optimal dalam mengekstraksi pola hubungan antar variabel secara mendalam, sedangkan model DL membutuhkan deret waktu yang panjang dan rentan terhadap overfitting pada data berukuran kecil. Penelitian ini mengusulkan model hybrid stacking ensemble untuk menggabungkan keunggulan kedua pendekatan tersebut melalui variasi antar arsitektur model. Model yang diusulkan mengintegrasikan XGBoost, LGBM, CatBoost, MLP, dan DNN sebagai base learner, dengan Logistic Regression (LR) dan Random Forest (RF) sebagai meta-model. Seleksi fitur menggunakan Mutual Information, pencarian hyperparameter dengan Optuna, serta uji statistik Friedman Rank dan Post Hoc Holm untuk mengukur perbedaan performa antar model. Hasil penelitian menunjukkan bahwa model hybrid stacking dengan meta-model RF mencapai akurasi 92.47%, presisi 92.81%, dan F1-score 89.65%, mengungguli model tunggal maupun stacking sejenis. Uji Friedman membuktikan adanya perbedaan performa antar model (p-value < 0,05), dan uji Post Hoc Holm mengonfirmasi bahwa model hybrid stacking-RF secara statistik lebih baik dibandingkan model DL serta model dengan konfigurasi default. Analisis lanjut dengan SHAP berhasil mengidentifikasi IPS Semester 3, Rata-Rata IPS, dan Skor TOEFL sebagai faktor paling dominan dalam memprediksi kelulusan tepat waktu.
===================================================================================================================================
Predicting on-time graduation for master's students faces challenges due to short study duration and limited historical data availability. Machine learning (ML) and deep learning (DL) models have been used separately, yet each has its own limitations. ML models are less capable of extracting in-depth relationships among variables, while DL models require long sequential data and are prone to overfitting when applied to small datasets. This study proposes a hybrid stacking ensemble model to combine the strengths of both approaches through variations across model architectures. The proposed model integrates XGBoost, LGBM, CatBoost, MLP, and DNN as base learners, with Logistic Regression (LR) and Random Forest (RF) as meta-models. Feature selection employs Mutual Information, hyperparameter tuning uses Optuna, and statistical testing applies Friedman Rank and Post Hoc Holm tests to measure performance differences among models. The results show that the hybrid stacking model with RF as the meta-model achieves 92.47% accuracy, 92.81% precision, and 89.65% F1-score, outperforming both single models and homogeneous stacking models. The Friedman test confirms performance differences across models (p-value < 0.05), and the Post Hoc Holm test further confirms that the hybrid stacking-RF model is statistically superior to DL models and default-configured models. Further analysis using SHAP identifies Semester 3 GPA, Average GPA, and TOEFL score as the most dominant factors in predicting on-time graduation.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Deep Learning, Hybrid Stacking Ensemble, Machine Learning, Mahasiswa Pascasarjana, Prediksi Kelulusan, Deep Learning, Graduation Prediction, Hybrid Stacking Ensemble, Machine Learning, Master Student
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. T Technology > T Technology (General) > T174 Technological forecasting T Technology > T Technology (General) > T57.5 Data Processing
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55101-(S2) Master Thesis
Depositing User:	Raihan Adam Handoyo Winarso
Date Deposited:	29 Jul 2026 01:33
Last Modified:	29 Jul 2026 01:33
URI:	http://repository.its.ac.id/id/eprint/139125

Actions (login required)

View Item