Tension with the street-space argument
The method treats streets and open spaces as places just as important as buildings, because Jacobs, Gehl and Whyte describe them as real sites of daily life. The results of research question 2 make this harder. The model finds 89.6% of building stays, but only 42.7% of street stays and 49.2% of open-space stays, and it mistakes both mostly for buildings (Confusion matrix).
The cause lies in how the model was fitted, not in weak GPS data. The share_type term gives a lower value to a candidate with many neighbours of the same type. Street stays have about 1.9 times as many same-type candidates as building stays (13.17 against 7.09 on average), because the map stores streets as many short pieces (Candidate density bias toward building).
Two explanations are still possible:
- The problem belongs to this street map. A map with fewer, longer street pieces would make the gap smaller.
- Long, repeated place types like streets are at a disadvantage in any method that counts nearby candidates.
None of the papers checked can decide between the two. Testing other ways of cutting the street map could (Future work).
A subsection of the Discussion in the thesis. Part of Error analysis.
Linked from Candidate density bias toward building · Error analysis · Future work · Motivation · Thesis summary