Table 2.

Signal processing steps applied to GPS data collected.

GPS DATA
Preprocessing	The timestamps from GPS data are extracted and used to split the raw data file into days to build an initial time vector.
Filtering and data interpolation	Each GPS position with a reported precision less than the specified GPS precision (gp) value is filtered and removed from the dataset. The minimum gp = 5 m. For each point in the continuous timeseries without a GPS position (because of filtering or no position), a zero-order hold interpolation is used to complete the series so that each missing value is reported as the last valid position.
Clusters and transit identification	A temporal cluster is created when time-consecutive positions are within close proximity to each other, over a specific window frame. Adjacent clusters are then merged to form a larger cluster. A rolling window (length, overlap) is used to identify temporal clusters. A window is considered to form a cluster if the (1-γ) th quantile distance from the median spatial center of the window is less than a specified radius (r). Window length = 300 s. Window overlap = 50% γ value = 0.3 Maximum distance radius (r) = 30 m.
Transit classification	For each transit between two clusters, the type of activity is identified. Positions comprised in a transit period are considered to be in a vehicle if the RMS speed over a 90s period ≥ 10 km/h. Positions in a transit period not considered to be in a vehicle are classified as on foot.
GPS time series ellipse modeling	The minimum span ellipse that can fit all of the positions of the dataset is computed using a minimum covariance estimator. The ellipse does not have to be centered on home, as it encompasses the whole dataset.