Classification: Picking a Category

Classification isn't limited to two categories. Binary classification picks between exactly two options, like spam or not spam, but multi-class classification sorts an input into one of many possible categories — like tagging a photo as 'cat,' 'dog,' 'bird,' or 'other.'

Under the hood, most classifiers don't just blurt out one answer — they estimate how likely the input belongs to each category, then pick whichever category comes out most likely. A spam filter might be 92% confident an email is spam and 8% confident it isn't; it reports 'spam' because that's the stronger signal.

The stakes of choosing the right category list show up clearly in something like a medical scan triage tool, which might sort scans into 'urgent review needed,' 'routine follow-up,' or 'no action needed.' If that category list is missing an important case, the model has no way to express it — it can only pick from what it was given.

A photo-tagging tool sorts every image into "landscape," "portrait," "food," or "other." What kind of task is this?

A classifier that estimates how likely an input belongs to each category, and returns the category with the highest likelihood, is choosing based on ____.

A classification model can only choose from the fixed list of categories it was trained on, even if the true answer falls outside that list.