573 / 2019-02-28 23:44:39

Discerning spatial-temporal differentiation of daily activities among different social groups by combining cell phone data and survey data

Spatial-temporal differentiation,Daily activity,Social justice,Small sample enlargement simulation,Machine learning

全体主题 > Big data in Urban Planning and Management

Spatial justice is an eternal issue for urban planning and policies. Beyond residential segregation and exclusion, activity-space-based studies provide seminal knowledge of how different social groups behave apart in their daily activities though door to door diary surveys. However, it’s impossible for scholars to send questionnaires to every corner of the city, this limits development of high-precision and targeted policies. Big data, like cell phone data, is now tracking activity trajectories of almost every people in the city while detailed socioeconomic attributes of the individuals are absent, this also restricts its application for varieties of social topics. This paper attempts to address those problems by merging survey data and big data in help of machine learning methods. Using survey data of 1004 residents in Guangzhou, we cluster the individuals into five different social groups according to their census register, income, education background, occupation type and other attributes reflecting their social classes. We further exam the residents’ daily activity patterns from four aspects, including geometric characteristics of their activity space, built environment around their travel chains, social environment around their travel chains and the way they organize their travel chains. Machine learning algorithms are employed to construct a relationship matrix from individuals’ gender, age, housing rent and daily activity patterns to different social groups though classify training. Drawing on cell phone data which also describes users’ gender, age, housing rent and activity trajectories, we investigate their daily activity patterns and get to distinguish which social group the user belongs respectively when referring to the relationship matrix found in machine learning. We finally simulate high resolution socioeconomic attributes of residents at a considerable credibility with the help of cell phone data, thus a full map of daily activities for different social groups is depicted. The results suggest obvious spatial and temporal variation in daily activities for people from different social groups, and there may exists activity segregation at specific time and place in Guangzhou. Mechanisms driving this progress and possible policy interventions are also discussed.