Pengembangan Model Hibrida Untuk Prediksi Masa Tunggu Lulusan Menggunakan ElasticNet dan LightGBM Studi Kasus Pada Universitas XYZ

Aditya, Wigananda Firdaus Putra (2026) Pengembangan Model Hibrida Untuk Prediksi Masa Tunggu Lulusan Menggunakan ElasticNet dan LightGBM Studi Kasus Pada Universitas XYZ. Masters thesis, UPN Veteran Jawa Timur.

	Text (Cover) 23066020021-Cover - Copy.pdf Download (4MB)
	Text (Bab1) 23066020021-Bab1.pdf Download (81kB)
	Text (Bab2) 23066020021-Bab2.pdf Restricted to Repository staff only until 27 February 2029. Download (349kB)
	Text (Bab3) 23066020021-Bab3.pdf Restricted to Repository staff only until 27 February 2029. Download (240kB)
	Text (Bab4) 23066020021-Bab4.pdf Restricted to Repository staff only until 27 February 2029. Download (209kB)
	Text (Bab5) 23066020021-Bab5.pdf Download (4kB)
	Text (Daftar Pustaka) 23066020021-Daftarpustaka.pdf Download (136kB)
	Text (Lampiran) 23066020021-Lampiran.pdf Restricted to Repository staff only until 27 February 2028. Download (69kB)

Abstract

Graduate waiting time to obtain a first job is an important indicator for higher education institutions. It is commonly used in accreditation assessments and key performance indicators (IKU). This study develops a hybrid stacking model to predict graduate waiting time by combining ElasticNet and LightGBM. The data were drawn from Universitas XYZ’s Tracer Study and academic records for graduates from 2020-2023 (N=493, after data cleaning, the final dataset consisted of N=449 observations. The predictors included time to degree completion, GPA, TOEFL score, SSKM, internship grade (KP), thesis grade (TA), study program, province, gender, job-search duration, and number of job applications. The entire workflow was implemented within a pipeline to reduce the risk of data leakage, incorporating missing-value imputation, numerical feature standardization, ordinal encoding for ordered variables, and one-hot encoding for nominal variables. Hyperparameters were tuned using RandomizedSearchCV. The stacking model was constructed with ElasticNet and LightGBM as base learners and ElasticNet as the meta-learner trained on out-of-fold predictions. Internal evaluation using cross-validation showed that the stacking model achieved the best and most stable performance with mean R²=0.5884, mean RMSE=2.0168, and mean MAE=1.6429. External testing on an independent 2024 graduate cohort (N=62) yielded R²=0.5304 with RMSE=2.5277 and MAE=1.9387, indicating that performance was largely maintained although errors increased for higher waiting-time cases that were relatively rare. Permutation importance highlighted several top features as dominant predictors. This model can be used as an analytical decision-support tool for mapping waiting-time and designing more targeted career service interventions.

Item Type:

Thesis (Masters)

Contributors:

Contribution	Contributors	NIDN/NIDK	Email
Thesis advisor	Agussalim, Agussalim	NIDN198508112019031005	agussalim.si@upnjatim.ac.id
Thesis advisor	Parlika, Rizky	NIDN0718058401	rizkyparlika.if@upnjatim.ac.id

Subjects:

Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Q Science > QA Mathematics > QA76.87 Neural computers

Divisions:

Faculty of Computer Science > Magister Information Technology

Depositing User:

Wigananda Firdaus

Date Deposited:

04 Mar 2026 02:54

Last Modified:

04 Mar 2026 02:54

URI:

https://repository.upnjatim.ac.id/id/eprint/49413