Detecting outlying demand in multi-leg bookings for transportation networks

Network effects complicate transport demand forecasting in general and outlier detection in particular. For example, a sudden increase in demand for a specific destination will not only affect the legs arriving at that destination, but also connected legs nearby in the network. Network effects are particularly strong when service providers, such as railway or coach companies, offer many multi-leg itineraries. In such situations, automated alerts can help analysts to adjust demand forecasts and enable reliable planning. In this presentation, we outline a novel two-step method for automatically detecting outlying demand from transportation network bookings. The first step clusters network legs according to the observed booking patterns. The second step identifies outliers within each cluster to create a ranked alert list of affected legs. We illustrate the method using empirical data obtained from Deutsche Bahn. In addition, we present a detailed simulation study that quantifies the improvement from the clustering step and implications of ranking to measure the criticality of the outliers. Our results show that the proposed approach outperforms independently analysing each leg, especially in highly connected networks where most passengers book multi-leg itineraries.

Keywords: Analytics; Forecasting; Outlier detection; Network analysis; Simulation; Demand management.