Error analysis
Part of Thesis summary. Research question 2: where the final model makes mistakes, and why.
- Confusion matrix: the model finds building stays about twice as often as street or open-space stays
- Candidate density bias toward building: street stays have the most street candidates around them, and the model punishes that
- Fold difficulty: the hard test sets are hard for every type, not only for streets
- Tension with the street-space argument: what this means for the idea that streets are real places of activity
- Candidate density as boundary-matching difficulty: a similar problem known from GPS research on roads and sidewalks
Meetings: RQ2 diagnostics and RQ3 ceiling check.