Time series clustering methods for categorizing active travel trends

Abstract

Active travel (AT) data has many uses in transportation planning, engineering, public health, and recreational planning. Often, direct measures of biking and walking are not available to transportation agencies, but proxy (i.e., indirect measures) of biking and walking are, which leads to interest in using them to inform understanding of AT trends. Our work investigates two topics that can direct future use of AT proxy data in transport problems: (1) we investigate the feasibility identifying travel typologies in proxy data sets; and (2) we examine three methods of time series clustering to assess each approach’s suitability for clustering AT proxy data. We apply these topics to two examples of AT data — self-reported bicycle data and pedestrian “push-button” data at intersections — and we compare the clusterings with qualitative and quantitative measures. Our work shows it is possible to extract typologies from AT proxy data, although the typologies are less distinct than they likely would be in true count data. We find that shape-based clustering results in cohesive, separated clusters that relate to socioeconomic and land use variables that are known to influence travel demand. In some cases, a simpler feature-based clustering produces high-quality clusterings on the bike data, providing practitioners with less complex options when applicable.