A Data-Driven Decision-Support Ticketing Prototype for Dynamic Pricing in the Bandung-Jakarta Intercity Transportation Corridor

Abstract

This paper reports the implemented analytical and predictive layers of a data-driven decision-support framework for dynamic pricing in the Bandung-Jakarta intercity corridor. The study follows Design Science Research Methodology and uses observational snapshots from KAI intercity rail, BHISA shuttle, and Whoosh high-speed rail. A provenance-preserving hourly data warehouse and OLAP layer first characterizes cross-modal repricing heterogeneity. KAI is then reconstructed at the physical-departure–class–observed-hour grain to model inventory trajectories rather than fare-subclass fragments. The predictive target, future_pressure_entry_t6, asks whether a currently non-critical KAI departure-class enters a critical state (remaining capacity ≤ 10%) within six hours, for prediction indices at 6–24 hours to departure. OBSERVED, causal-safe MEAN, MEDIAN, and KNN variants are evaluated using matched cohorts, grouped chronological splits, and a sealed one-shot test. C4.5 achieved the highest validation AUPRC (0.629) among C4.5, CART, Random Forest, and Naive Bayes. On the final OBSERVED reference test, AUPRC was 0.391 and ROC-AUC 0.723. Missing-data treatments produced metric-dependent behavior on the matched resolved cohort. The resulting score is interpreted as an early-warning risk signal, not as a pricing action; downstream pricing-policy implementation is outside the empirical scope of this paper.