Detection of base travel groups with different sensitivities to new high-speed rail services: Non-negative tensor decomposition approach

Abstract How many base travel groups (models) are necessary for clarifying the long-term day-to-day dynamics of intercity travel? In the past, several travel purposes (e.g., sightseeing, business, etc.) have been assumed. However, mobile-phone location data enables us to answer the above question because of their detailed time-series information. In this study, we propose a method for deriving the basic travel groups necessary for clarifying the time-series changes by applying nonnegative tensor factorization (NTF). This method is applied to the time-series data of several long-distance travelers to the Ishikawa prefecture, to where the Hokuriku High-speed rail (HSR) has been newly extended. Based on this, the number of base travel groups necessary for predicting the effect of the new HSR is estimated as twelve, which is greater than the number used in the previous demand forecasting models. The estimated groups include components that appear to correspond to different travel purposes (e.g., sightseeing, business, and homecoming), as in previous surveys. These results confirm that the methodology proposed in this study can clearly extract groups with different elasticities, due to the traffic service. The HSR effect can be clarified by dividing it into several characteristics and detailed components. In addition, if multiple HSR effects are analyzed, a more accurate demand-forecasting model for the new HSR service can be proposed.