Application of Clustering Method for Segmentation of Traffic Accident Profiling With K-Means And K-Medoids Case Study of Toll Bakauheni – Terbanggi Besar

Abstract

Traffic accidents are a serious problem that requires in-depth analysis to identify patterns and risk factors that influence them. This study applies the clustering method in data mining techniques to segment traffic accident profiling on the Bakauheni - Terbanggi Besar Toll Road. Two clustering algorithms, namely K-Means and K-Medoids, are used to divide accident data into three zones, namely the safe zone, the alert zone, and the danger zone. The main objectives of this study are to identify accident distribution patterns and evaluate the performance of the two clustering methods. The results of clustering using the K-Means method show that 29.7% of the data are classified as the danger zone, 38.9% are in the safe zone, and 31.4% are in the alert zone. Meanwhile, the results of clustering using K-Medoids show that 32.8% of the data are in the danger zone, 34.2% are in the safe zone, and 33.1% are in the alert zone. To evaluate the quality of the clustering results, testing was carried out using three evaluation metrics, namely Silhouette Score, Davies-Bouldin Index (DBI), and Purity. In the K-Means method, the Silhouette Score was obtained at 0.365, DBI at 1.117, and Purity at 0.617. While in the K-Medoids method, the Silhouette Score was obtained at 0.273, DBI at 1.440, and Purity at 0.436. Based on the evaluation results, the K-Means method showed better clustering performance than K-Medoids, with a higher Silhouette Score value and a lower DBI, indicating better cluster quality.