Understanding Activity Location Choice with Mobile Phone Data

Abstract

Abstract. Research on the variability in travel behavior, i.e. how individuals repeat or change their travel behavior over time, has a long history. An understanding of variability not only facilitates behavior modeling efforts but also provides insights into the complex factors underpinning people's travel behaviors. Previous studies of behavioral variability have mostly concentrated on non-spatial aspects. This dissertation contributes to the existing literature by examining the spatial variability of activity location choices. Activity locations are a set of spatially dispersed places where individuals perform activities. They play a critical role in structuring human travels, as individuals' demand for travelling in space is derived from the demand for activity participation. Specifically, two research questions are of interest: 1) how is location variability affected by time-of-day? 2) how can the knowledge of location variability be used to inform the development of location prediction models? Analyses are performed with a mobile phone data set consisting of the traces of 120,435 individuals over two months. Significant time-of-day dependence of location variability is identified. Time-of-day effect is found to take account for 36% of the total variations in location variability. The results emphasize the importance of time-of-day in shaping one's location choice behavior and provide a basis for future modeling efforts on location variability. Location variability is found to be an instrumental indicator of the amount of input information required for location prediction. Specifically, given 100 historical (not necessarily unique) locations, an accuracy level marginally over 80% can be achieved for people with low location variability. In contrast, for those individuals with a high level of location variability, prediction accuracy can hardly reach 50% with 100 historical locations. This finding has significant implications on making more efficient location predictions. It will allow us to customize the amount of information input in location prediction for subpopulations differing in location variability by removing redundant information. Being able to discard some data without compromising model prediction accuracy is one way to deal with an overwhelming amount of data.