Exam-style
A marketing team has three years of subscriber behaviour: box sizes, delivery frequency, substitutions accepted, pauses taken, spend per month. No column records which group a subscriber belongs to, and nobody has ever written such a group down. They want to know whether there are natural groups worth treating differently, and they cannot say in advance how many there might be or what to call them. Which technique fits?
Reveal the answer
C. Clustering, an unsupervised method that groups subscribers by similarity with no target column to predict
Two facts about the data settle this before any algorithm is named. No column records the segment, so there is nothing for a model to predict. There is no fixed set of categories either, so nothing to predict into. Both halves of supervised learning are missing, which leaves unsupervised learning, where the method that proposes groups is clustering. Classification, multi-class or binary, needs the categories defined in advance and labelled examples of each; here the categories are the thing being looked for. Regression fails differently. Monthly spend is in the data, so a spend model trains fine; it predicts a number when a grouping is wanted, and banding those predictions draws the groups along one axis somebody chose. Semi-supervised learning is the near miss. It fits when the categories are already known and the labelled examples are scarce. Here, whoever labels the sample has to invent the segments first, which imposes the categories instead of finding them. Two caveats. With k-means, how many groups to look for is a number you set and try several values of, not an output of the run. And the clusters come back unnamed: reading the centres and deciding that group three is the fortnightly bulk buyers is human work.
Q. Three years of subscriber behaviour, no labels, no target column, and nobody knows how many groups there are or what to call them. Which technique?
A. Clustering, which is unsupervised learning: it groups records by similarity with nothing to predict.
Why? Classification needs a fixed set of labels agreed in advance and labelled examples of each, and this data has neither: the categories are the thing being looked for. Regression is trainable here, since monthly spend is in the data. But it predicts a number rather than a grouping, and banding those predictions draws the segments along one axis somebody chose. Semi-supervised learning is the near miss. It fits when the categories are known and the labels are thin on the ground. Labelling a sample here means inventing the segments before you look for them, so the groups that come back are the ones the labeller made up.