Research questions
Goal: a method that gives each GPS stay a probability for building, street and open space. The method takes GPS error into account by measuring each stay itself: how spread out its points are, and how they move over time.
- RQ1: How can GPS error be built into a method that gives probabilities for several types of place? → Framework overview, and how the model was built step by step: Calibration
- RQ2: How well do the method's answers match the correct answers from the field survey? → Confusion matrix
- RQ3: Does a more complex model fit better? And does that extra complexity make it worse on new data, where the mix of place types may be different? → Overfitting signature, ML ceiling check, row level
Part of Problem and positioning.
Linked from Confusion matrix · Framework overview · Motivation · Overfitting signature · Problem and positioning · Thesis summary