Limitations
What the results of this thesis cannot show.
- Planned behaviour. Participants followed instructions, so their behaviour was not fully natural. This matters most for the stops with small movements (Group D: movement type).
- Three phones per stop. The three recordings of one stop are not independent. Grouped cross-validation reduces this problem but does not remove it.
- Terraces and balconies. "Building-attached Outdoor" is counted as open space (Self-report form).
- One street map. The street results come partly from how this map cuts streets into pieces, so the street and open-space numbers belong to this map (Candidate density bias toward building).
- Stays one by one. Each stay is judged alone, without the trips before and after it.
- No MILES test yet. The model has not been used on MILES, so the risk described in Overfitting signature has not been tested.
Part of Outlook.
Linked from Future work · Outlook · Overfitting signature · PhD interview slides · Self-report form · Study area