A graphical representation of the alternatives in a decision-making problem. As an example , suppose that we are choosing between two machines. One costs 100 units and has a 20% probability of breaking down within a year. The other costs 120 units and has a 5% breakdown probability. A breakdown costs 60 units. Ignoring the possibility of multiple breakdowns, which machine should we buy? Reading the decision tree from right to left, we see that it is cheaper to buy the 100-unit machine.

In machine learning the ‘decisions’ are, in effect, the answers to questions, with the final values at the ends of the tree being predicted values or classes. Pruning is a technique for simplifying such trees by removing sections that have minimal impact on the final decision or value.

Decision tree. A circle at the right contains the total cost if this path occurs. An intermediate circle gives the expected costs at this stage. Thus, for example, 112=100×80% + 160×20%. If cost is the only consideration, then in the case illustrated the cheaper machine is the better buy.