Search

Utilization of movement data from mobile sports tracking applications is affected by its inherent biases and sensitivity, which need to be understood when developing value-added services for, e.g., application users and city planners. The authors have developed a method for generating a privacy-preserving heat map with user diversity (ppDIV), in which the density of trajectories, as well as the diversity of users, is taken into account, thus preventing the bias effects caused by participation inequality. The method is applied to public cycling workouts and compared with privacy-preserving kernel density estimation (ppKDE) focusing only on the density of the recorded trajectories and privacy-preserving user count calculation (ppUCC), which is similar to the quadrat-count of individual application users. An awareness of privacy was introduced to all methods as a data pre-processing step following the principle of k-Anonymity. Calibration results for the authors' heat maps using bicycle counting data gathered by the city of Helsinki are good (R2 > 0.7) and raise high expectations for utilizing heat maps in a city planning context. This is further supported by the diurnal distribution of the workouts indicating that, in addition to sports-oriented cyclists, many utilitarian cyclists are tracking their commutes. However, sports tracking data can only enrich official in-situ counts with its high spatio-temporal resolution and coverage, not replace them.

Crowd-sourced data of high spatial and temporal resolution can provide a new basis for mobility analyses given that its various types of biases distorting the results are identified and adequately handled. In this paper, trajectory patterns that can affect the validity of mobile fitness app data are examined by means of cycling trajectories (n = 50,524) from the Helsinki Metropolitan Area, in Finland. In addition to mass events and group journeys, we evaluated the biasing effect of routes that have been repeatedly recorded by the same application user. Based on the results, repeatedly recorded commuting routes may skew fitness application data more than group patterns. Many of the changes in the frequencies and length distributions at different temporal granularities before and after extracting the ‘bias patterns’ were statistically significant. Also the skewed distribution of tracks among users (i.e. contribution inequality) became more even. The biases induced by behavioural patterns ought to be considered when evaluating the validity of fitness app data in analyses of general mobility behaviour and when designing value-added applications based on the data. Considering the trade-off between privacy and data accuracy regarding dissemination of sensitive crowd-sourced movement data, the findings emphasise the importance of preserving the possibility to detect individual-level phenomena in order to produce valid analysis results.