Import Logistics Performance Optimization through Duration-Based Process Clustering Using K-Means

Abstrak

In the global logistics landscape, optimizing import operations is critical to ensuring timely delivery and minimizing inefficiencies. This study proposes a hybrid clustering framework that integrates K-Means with Particle Swarm Optimization (PSO) to classify import logistics processes based on duration metrics. A data set comprising 344 real-world import records was analyzed, with pre-processing steps including RobustScaler normalization and outlier handling to improve data quality. The PSO algorithm was used to dynamically optimize the clustering parameters, leading to better segmentation of the logistics stages into performance-based clusters. Internal validation using the Silhouette Score (0.987), Davies-Bouldin Index (0.01), and Calinski-Harabasz Index (2450) confirmed the superior performance of K-Means over DBSCAN and Agglomerative Clustering. Principal Component Analysis (PCA) further visualized the separation between fast and slow process groups. The findings reveal that the delivery stage represents the most significant bottleneck, with durations exceeding 700 days in slower clusters. The proposed method offers actionable insights for logistics managers to improve operational efficiency, reduce lead time variability, and implement data-driven process improvements.