Evaluasi Performa Algoritma Klasterisasi dalam Mengelompokkan Topik Tugas Akhir: Studi Komparasi K-Means dan DBSCAN

Abstract

The increasing number of Informatics Engineering students at STT-NF has resulted in a growing volume of final project topics that need to be efficiently categorized into research areas. This study evaluates the performance of the K-Means and DBSCAN algorithms for clustering students' final project topics based on their textual characteristics. The research follows the CRISP-DM framework and utilizes a dataset of 656 final project titles. The data were processed through text preprocessing, TF-IDF weighting, and dimensionality reduction PCA. The experimental results indicate that K-Means with k = 13 achieved the best performance, obtaining a silhouette score of 0.1434 and a purity score of 0.8659, outperforming DBSCAN, which achieved a silhouette score of 0.1153 and a purity score of 0.6589. The resulting clusters were successfully mapped into four major research areas, namely Software Engineering, Network Engineering & Cyber Security, Data Engineering, and UI/UX Design. Furthermore, the best-performing model was implemented in an interactive Streamlit-based dashboard to support research area mapping and academic supervisor assignment.

Keywords: K-Means; DBSCAN; Final Project Topic Clustering; TF-IDF

Abstrak

Peningkatan jumlah mahasiswa Teknik Informatika di STT-NF menghasilkan semakin banyak data topik tugas akhir yang perlu dikelompokkan ke dalam bidang penelitian secara efisien. Penelitian ini mengevaluasi performa algoritma K-Means dan DBSCAN untuk mengelompokkan topik tugas akhir mahasiswa berdasarkan kemiripan karakteristiknya. Proses penelitian mengikuti metodologi CRISP-DM dengan menggunakan 656 judul tugas akhir yang diproses melalui tahapan pemrosesan teks, pembobotan TF-IDF, dan reduksi dimensi PCA. Berdasarkan hasil pengujian, K-Means dengan k = 13 memberikan kinerja terbaik dengan nilai silhouette score 0.1434 dan purity 0.8659, sedangkan DBSCAN memperoleh nilai silhouette score 0.1153 dan purity 0.6589. Klaster yang terbentuk berhasil direpresentasikan ke dalam empat bidang penelitian utama, yaitu Software Engineering, Network Engineering & Cyber Security, Data Engineering, dan UI/UX Design. Sebagai implementasi, model terbaik diintegrasikan ke dalam dashboard interaktif berbasis Streamlit untuk mendukung pemetaan bidang penelitian dan penentuan dosen pembimbing.

Kata kunci: K-Means; DBSCAN; Klasterisasi Topik Tugas Akhir; TF-IDF