<![CDATA[The Jabodetabek Commuter Rail (KRL) serves millions of passengers daily across a network of 94 active stations and 421 direct service relations. In highly interconnected networks, certain stations are critical, defined as those whose disruption causes the greatest impact on network connectivity and average travel time efficiency. This study develops a predictive model based on Node2Vec and LightGBM under the Knowledge Discovery in Databases (KDD) framework to numerically estimate station criticality levels. Operational schedule data were collected via the Comuline API (30,989 records), processed into a directed weighted graph (94 nodes, 421 edges), and used to compute impact scores through single-node removal simulation as regression labels. A total of 38 features were used: 6 topology metrics and 32 Node2Vec embedding dimensions. 5-Fold Cross-Validation yielded MAE 0.3152%, RMSE 0.8903%, and Spearman correlation 0.7802 (p<0.001) with 95% Bootstrap CI [0.6622; 0.8700]. The model identified 9 of 10 most critical stations (Top-10 recall 90%). Betweenness centrality was the most dominant predictor based on Gain Importance and SHAP analysis. Of 94 stations, 8 are Very Critical, 21 Critical, 20 Moderately Critical, and 45 Relatively Safe. Manggarai and Tanah Abang are the most critical with impact scores of 9.60% and 8.89%.]]>
Copyrights © 2026
