Prediksi Kelulusan Pada Data Histori Mahasiswa Menggunakan Deep Learning Dan Parallel Computing

Damara, Rule Lulu (2025) Prediksi Kelulusan Pada Data Histori Mahasiswa Menggunakan Deep Learning Dan Parallel Computing. Other thesis, Institut Teknologi Sepuluh Nopember.

	Text 5025211050-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (8MB) | Request a copy
	Text 5025211050-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (8MB) | Request a copy

Abstract

Kelulusan tepat waktu mahasiswa merupakan indikator penting dalam penilaian akreditasi program studi dan kualitas pendidikan tinggi. Penelitian ini bertujuan mengembangkan model prediksi kelulusan mahasiswa menggunakan teknik deep learning yaitu DNN, CNN, dan LSTM dengan memanfaatkan parallel computing untuk meningkatkan efisiensi komputasi. Sistem dirancang untuk melakukan prediksi pada setiap semester menggunakan data historis mahasiswa yang sudah lulus dari tahun 2004-2023. Data telah dipisahkan berdasarkan jenis institusi (PTN, PTS, dan Kedinasan) dan dibagi per semester secara kumulatif, artinya data semester ke-n mengandung informasi akademik mahasiswa dari semester 1 hingga semester n. Dataset yang terdiri dari 306.703 baris data dengan 48 kolom fitur dan 1 kolom target serta mengalami ketidakseimbangan kelas. Oleh karena itu digunakan teknik SMOTE untuk menyeimbangkan distribusi kelas. Pendekatan berbasis semester memungkinkan sistem memberikan prediksi dini pada setiap semester akademik, sehingga institusi pendidikan dapat melakukan intervensi preventif lebih awal. Proses pelatihan model dilakukan secara paralel menggunakan modul concurrent.futures Python untuk mengoptimalkan waktu komputasi. Evaluasi model menggunakan metrik precision dan recall guna meminimalkan risiko false positive dan false negative. Hasil penelitian menunjukkan bahwa seleksi fitur menurunkan akurasi 3-7% namun tetap dipertahankan untuk efisiensi komputasi, sementara SMOTE efektif meningkatkan recall kelas minoritas. Hyperparameter tuning mengungkap konfigurasi optimal yang berbeda setiap institusi, dengan DNN sebagai model terunggul yaitu akurasi tertinggi 81.51% pada dataset Kedinasan, dataset PTS 76.57%, dan dataset PTN 70.54%. Implementasi parallel computing mampu meningkatkan efisiensi komputasi hingga 65.6%.
==================================================================================================================================
Timely student graduation is an important indicator in study program accreditation assessment and higher education quality. This research aims to develop student graduation prediction models using deep learning techniques including DNN, CNN, dan LSTM by leveraging parallel computing to improve computational efficiency. The system is designed to provide predictions at each semester using historical data of graduated student from 2004 to 2023, which has been separated based on the type of institution (PTN, PTS, and Kedinasan) and divided cumulatively by semester, where the dataset for the n-th semester contains academic information from semester 1 through semester n. The dataset consists of 306,703 rows with 48 feature columns and 1 target column, and it exhibits significant class imbalance. Therefore, the SMOTE method was applied to balance the class distribution. This semester-based approach allows the system to provide early predictions at each academic stage, enabling educational institutions to implement preventive interventions earlier. The model training process was executed in parallel using Python’s concurrent.futures module to optimize computation time. Model evaluation employed precision and recall metrics to minimize false positive and false negative risks. The results indicate that feature selection reduced accuracy by 3-7% but was retained for computational efficiency, while SMOTE effectively improved minority class recall. Hyperparameter tuning revealed institution-specific optimal configurations, with DNN emerging as the top-performing model, achieving the highest accuracy of 81.51% for Kedinasan, 76.57% for PTS, and 70.54% for PTN. Parallel computing implementation improved computational efficiency by up to 65.6%.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	CNN, DNN, Deep Learning, LSTM, Parallel Computing, Prediksi Kelulusan CNN, Deep Learning, DNN, Graduation Prediction, LSTM, Parallel Computing
Subjects:	T Technology > T Technology (General) > T58.62 Decision support systems
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Rule Lulu Damara
Date Deposited:	25 Jul 2025 03:22
Last Modified:	25 Jul 2025 03:22
URI:	http://repository.its.ac.id/id/eprint/121156

Actions (login required)

View Item