PUTRO, Rochsid Tri Hanggoro and Warsito, Budi and Surarso, Bayu (2026) SEGMENTASI MAHASISWA BERDASARKAN PERILAKU PADA LEARNING MANAGEMENT SYSTEM DENGAN TWIN SUPPORT VECTOR MACHINES. Masters thesis, UNIVERSITAS DIPONEGORO.

	Text 1. cover awal.pdf Download (310kB)
	Text 2. cover lengkap.pdf Restricted to Repository staff only Download (4MB)
	Text 3. BAB I.pdf Download (295kB)
	Text 4. BAB II.pdf Download (932kB)
	Text 5. BAB III.pdf Restricted to Repository staff only Download (633kB)
	Text 6. BAB IV.pdf Restricted to Repository staff only Download (2MB)
	Text 7. BAB V.pdf Restricted to Repository staff only Download (285kB)
	Text 8. Daftar Pustaka.pdf Download (259kB)

Abstract

Learning Management System (LMS) merupakan jejak aktivitas digital mahasiswa yang dapat dianalisis, sehingga mendukung pemahaman tentang segmentasi pola belajar. Namun, segmentasi perilaku sering kali bermasalah karena ketidaksediaan label perilaku secara eksplisit. Tujuan penelitian ini adalah mengembangkan pendekatan segmentasi dengan klasifikasi berbasis data log LMS mahasiswa dengan skema weakly supervised melalui pseudo-labeling, serta membangun model klasifikasi hierarkis untuk segmentasi mahasiswa ke dalam empat segmen yaitu Consistent, Crammer, Passive, dan At-Risk. Data pada penelitian ini adalah log aktivitas pembelajaran LMS berupa forum_view_count, login_count, submission_ratio, assignment_open_lag, submission_lead_time, activity_count, dan session_gap_hours dari 32.146 mahasiswa dalam satu semester. Pra-pemrosesan pada data menggunakan imputasi nilai hilang, winsorization untuk mengatasi outlier, dan normalisasi fitur terdistribusi antara [0–1]. Pseudo-label dibuat menggunakan K-Means dengan k=4 sebagai inisialisasi label dan sistem melakukan klasifikasi menggunakan Hierarchical Twin SVM dengan tiga node yaitu Active vs Inactive, Consistent vs Crammer pada Active, Passive vs At-Risk pada Inactive. Pengujian dengan stratified split 70:30 diperoleh akurasi keseluruhan 84,2% dengan Cohen’s Kappa 0,782, sehingga sudah valid. Metode yang digunakan efektif untuk membedakan segmen pola perilaku mahasiswa untuk intervensi akademik yang lebih tepat.
Kata kunci: Learning Analytics, LMS, Segmentasi Mahasiswa, K-Means, Twin SVM, Klasifikasi Hierarkis

The LMS contains a log of students' digital activities, which can be used to understand better the different segments into which various learner patterns can be placed. Nonetheless, behavior segmentation remains a challenging task since there are no explicit behavior inventories. This research proposal will also develop a segmentation method that can accurately segment student LMS log data by leveraging classification and weak supervision of pseudo-labeling, establishing a hierarchical classification model to identify students into four groups: Consistent, Crammer, Passive, and At-Risk. The dataset includes the following features, based on one semester of LMS activity logs from 32,146 students: forum_view_count, login_count, submission_ratio, assignment_open_lag, submission_lead_time, activity_count, and session_gap_hours. Data-hot-processing includes missing-value imputation, outlierness winsorization, and feature normalization to the [0–1] interval. Pseudo-labeling: The pseudo-label is created by applying K-Means (k = 4) as the first step for labeling, and then the system iteratively classifies using Hierarchical Twin SVM with three binary decision nodes: Active vs. Inactive, Consistent vs. Crammer (among active), and Passive vs. At-Risk (within Inactive). With a stratified 70:30 split, the model achieves an accuracy of 84.2%, and there is an agreement of κ = 0.782 between raters in the very good segmentations obtained. This method successfully identifies patterns of behavioral gain in students with implications for more tailored academic interventions.
Keywords: Learning Analytics, LMS, Students Segmentation, K-Means, Twin SVM, Hierarchical Classification

Actions (login required)

View Item