Masters candidate in Biostatistics, Siyang Ren, will present:
“Exploring Daily Activity Patterns from Mobile Phone Data”
Plan B Adviser: Julian Wolfson
Abstract: Smartphone-based activity surveys allow the collection of detailed information about individuals’ trips and activities in near real-time. However, the multidimensional nature of these data makes it challenging to compare activity-travel behaviors and identify clusters of individuals with similar behavior patterns. In contrast to previous studies on this topic, we adopted a discrete view on time by segmenting episodic activity-travel data into short, equal-length intervals labeled with a single character indicating the activity or travel type during that interval. The resulting “travelomes” were used to characterize individual daily and weekly activity-travel patterns. Motivated by techniques in genomics, we applied weighted sequence alignment algorithms to assess the pairwise similarity between individuals’ travelomes. To account for flexibility of trips and activities, we investigated several sequence alignment weighting schemes, for example by varying the relative substitution penalty for fixed and flexible activity types. The resulting distance matrices can be used to assess the activity-travel pattern similarity between pre-defined groups (e.g., men and women) or as inputs to a hierarchical clustering algorithm to identify clusters of individuals with similar travelomes. To compare the difference in activity-travel patterns between groups, we applied functional data analysis. We treated each individual’s daily activity as a continuous process, and the travelomes as a sample from it. We used B-spline basis functions to estimate this continuous process. The estimated functional curve was then fitted in a functional linear model as response variable. The coefficients estimated from this model describe the characteristic of each behavior pattern. A subsequent functional analysis of variance (fANOVA) was performed on the regression curves to decide when and how these activity-travel patterns are significantly different with each other. We applied our method to data collected by a smartphone application, Daynamica, on 372 individuals in the Twin Cities metropolitan area. We showed that the activity-travel patterns are affected by reasons such as gender, spouse and whether they have children younger than age 18. We also identify five common behavioral patterns in weekdays and weekends. The most common weekdays pattern is defined as “Home-work” pattern and they are also some sub-patterns in people who staying at home for most of the day. This pattern is also pretty common on weekends. The results of fANOVA shows that in the morning, the difference between groups’ patterns are insignificant, while in other time intervals, the difference is quite large.
Refreshments will be served prior to the presentation.