Decision tree
A decision tree is a branching model that reaches a decision or prediction through a sequence of questions, each splitting the data or the options into branches. In analytics it is a supervised learning method for classification and regression.
How it works
A tree starts at a root node, asks a question about one variable (is monthly spend over $50?), and follows a branch for each answer to an internal node, which asks the next question. The path ends at a leaf node that gives the outcome, such as "likely to renew". Learning algorithms choose each split to separate the outcomes as cleanly as possible, using measures such as Gini impurity or information gain.
Trees are easy to read and explain, which makes them popular for segmentation and churn analysis. They can overfit, memorizing noise in the training data, so analysts limit their depth or combine many trees into a random forest.
A worked example
Trained on 8,000 renewal records, a subscription software company's decision tree splits first on seats used. Accounts using under 40 percent of their seats renew at 52 percent, against 91 percent above. The next split, support tickets in the last quarter, identifies a group of 600 accounts with a 34 percent renewal rate.
Related
Automate this in Pulse
The first split a decision tree makes, the variable that best separates buyers from browsers, can become the screener question of a Pulse brief, so interviews hear from people on each side of it.