<![CDATA[The rapid growth of urban populations and increasing reliance on public transportation in Indonesia present challenges in managing passenger demand effectively. In Surabaya, the steady rise in Suroboyo Bus passengers underscores the need for data-driven strategies to optimize fleet allocation, scheduling, and infrastructure development. Identifying passenger density patterns through clustering provides a systematic basis for decision-making. This study aims to address a local research gap by comparing three clustering algorithms Agglomerative Hierarchical Clustering (AHC), Gaussian Mixture Model (GMM), and K-Medoids on empirical passenger data. Unlike previous studies that emphasize route optimization or demand forecasting, this research highlights a comparative evaluation to determine the most effective method for handling fluctuating and outlier-prone transportation data. The dataset was obtained from the Surabaya City Transportation Office for the Purabaya–Perak route during a two-week period in 2024. Data preprocessing included attribute selection, transformation of time into numerical format, outlier detection using the Interquartile Range (IQR), and Z-Score normalization. Clustering results were assessed with the Silhouette Score and visualized using scatter plots and histograms. Findings show that K-Medoids achieved the highest Silhouette Score (0.4222), surpassing AHC (0.3657) and GMM (0.3024). K-Medoids produced more balanced clusters and stronger resilience to outliers, while AHC provided interpretable hierarchical structures, and GMM modeled complex patterns but with weaker separation. In conclusion, K-Medoids is recommended as the most suitable approach for passenger density clustering. Academically, this study contributes a comparative framework for clustering in transportation research, while practically offering insights to support data-driven public transport management in developing cities.]]>
Copyrights © 2025
