Analisis Perbandingan Algoritma K-Means dan DBSCAN Untuk Segmentasi Pemain Mobile Legends Berdasarkan Performa dan Pola Bermain di Wilayah Medan

Abstract

Player segmentation in Mobile Legends: Bang Bang is important to understand the variations in performance and playing styles in local communities such as Medan. This study aims to compare K-Means and DBSCAN algorithms in clustering players based on authentic data from 245 active players in Medan collected through surveys in January-March 2025. The data was analyzed using eight clustering evaluation metrics, including Silhouette Score, Davies-Bouldin Index, Calinski-Harabasz Index, Adjusted Rand Index, as well as ANOVA and Kruskal-Wallis statistical tests. The results showed that K-Means produced 4 clusters with Silhouette Score 0.569 and Davies-Bouldin Index 0.891, while DBSCAN produced 9 clusters with Silhouette Score 0.659 and detected 5 outlier players (2.04%). Four player archetypes were successfully identified through K-Means, while DBSCAN produced a more granular role-based segmentation with superior noise detection capabilities. This research also produced a Flask-based web interface as a real-time clustering analysis visualization tool with a draft pick recommendation system for 129 hero characters.