Exploring time diaries using semi-automated activity pattern extraction

Abstract

Identifying patterns of activities within individuals � time diariesand studying similarities and deviations between individuals in apopulation is of interest in time use research. So far, activitypatterns in a population have mostly been studied either by visualinspection, searching for occurrences of specific activity sequencesand studying their distribution in the population, or statisticalmethods such as time series analysis in order to analyse daily behaviour.We describe a new approach for extracting activity patterns fromtime diaries that uses, instead, sequential data mining techniques.We have implemented an algorithm that searches the time diaries andautomatically extracts all activity patterns meeting user-definedcriteria of what constitutes a valid pattern of interest for theresearch question. Amongst the many criteria which can be appliedare: a time window containing the pattern, and minimum and maximumnumber of people that perform the pattern. The extracted activitypatterns can then be interactively filtered, visualized and analyzedto reveal interesting insights using the VISUAL-TimePAcTS application.To demonstrate the value of this approach we consider and discusssequential activity patterns at a population level, from a singleday perspective, with focus on the activity �paid work � and someactivities surrounding it. Questions can be posed such as: Whichactivities appear frequently in activity patterns related to paidwork? Which are the activities surrounding work at different hoursof the day? What differences are revealed between the sexes in thepatterns? Exploration of the results of each pattern search may resultin new hypotheses which can be subsequently explored by alteringthe search criteria.