Estimating destination of unlinked trips in public transportation smart card fare collection systems

Smart card automated fare collection (SCAFC) systems have proved effective in collecting data on the travel behavior of users of public transit networks. Since some systems record only the boarding locations, we need a method for estimating destinations. Existing algorithms can find an estimate for most trips. However, unlinked trips, which are trips that are not part of a trip chain during the day, are more difficult to analyze. We present an improvement to the existing model for estimating destinations, in particular for unlinked trips. The method is based on kernel density estimation of the spatial and temporal probabilities of each destination. The data used for this study comes from the Société de transport de l’Outaouais, a mid-size bus service near Ottawa, Canada (for a one-month period in 2009 with 908,303 transactions). Existing algorithms can handle only 80.64% of the trips; our method handles an additional 10.9%. We present an analysis of these results and discuss future research directions.