Usability of a Wi-Fi fingerprint approach for place departure recognition in travel surveys
Smartphone travel behavior surveys, nowadays primarily based on location abilities of GPS, recently started to experiment with other available sensors to improve the quality of collected data. Wi-Fi “fingerprint” approach promises to supply GPS in place detection task in densely built areas and enclosed buildings. The main goal of the study is to verify the reliability of fingerprint approach to detect spatially defined place and its departure on the precision level needed in travel surveys. The study is based on the data (457 respondents and 606 places) from GPS pilot travel behavior survey conducted in the Czech Republic in 2013–2014. Using statistics of Informedness, Markedness and Matthews correlation coefficient we examined the ability of the three most common similarity measures (Jaccard, Tanimoto, Cosine) to correctly differentiate records from place (circle of 100 m radius) and non-place (concentric annulus with a 350 m radius). We found, there is an overall 3/5 chance of all respective cases of being retrieved (Informedness) and a 4/5 chance that the cases we retrieve are marked correctly (Markedness). At best we can explain less than half of the variance in the distribution of places/non-places.