Detecting outlying demand in multi-leg bookings for transportation networks

摘要: Network effects complicate demand forecasting in general, and outlier detection particular. For example, transportation networks, sudden increases for a specific destination will not only affect the legs arriving at that destination, but also connected nearby network. are particularly relevant when transport service providers, such as railway or coach companies, offer many multi-leg itineraries. In this paper, we present novel method generating automated alerts, to support analysts adjusting forecasts accordingly reliable planning. To create propose two-step detecting outlying from network bookings. The first step clusters appropriately partition pool booking patterns. second identifies outliers within each cluster ranked alert list of affected legs. We show outperforms analyses independently consider leg network, especially highly-connected networks where most passengers book illustrate applicability on empirical data obtained Deutsche Bahn with detailed simulation study. latter demonstrates robustness approach quantifies potential revenue benefits networks.