Why were decision boundaries visualized in 2D using PCA in the practical exercises?

A model performs very well on training data but poorly on test data. This is an example of:

In the NO₂ exceedance classification task, what does a false positive represent?

Which metric is most appropriate if the primary goal is to detect as many true Spruce/Fir habitats as possible?

In the habitat detection task, which misclassification is typically considered more environmentally risky?

Why was the original multi-class forest cover problem reformulated into a binary classification task?