An internal node tests a feature and sends observations along branches according to the test’s outcome. Repeated splits partition the data into progressively more homogeneous groups; leaves hold the final predictions. section 5.3.1
Choosing a split
A tree might test income rather than credit history, or compare income against a threshold. The selected test determines how observations divide at that node. section 5.3.1
The entropy calculation illustrates a split reducing uncertainty by 0.20 bits. section 5.3.1



