A decision tree is a predictive model that organizes feature-based conditional rules into a hierarchy. Each internal node splits transactions at a feature threshold, and each leaf predicts a class. section 3.4 section 3.4.4
A practical fraud-detection workflow uses a tree to:
- Split processed transaction data into training and test sets. section 3.4
- Train with cross-validation and randomized hyperparameter search. section 3.4
- Evaluate predictions with area under the precision-recall curve for imbalanced fraud data. section 3.4 section 3.9
- Interpret decisions by inspecting feature importance, split thresholds, node impurity, sample counts, and class votes. section 3.4.4
Reading a fraud decision tree
A root node can compare TRANSFER with a threshold of 0.5. Transfer transactions follow one subtree and other transactions follow the other; subsequent splits lead to leaves labeled fraudulent or genuine by the majority class. Gini impurity indicates how mixed the transactions are at each node. section 3.4.4
Figure 3.11 Steps in building a decision-tree-based online transactions fraud detection model.
Use the workflow figure to trace the progression from transaction data through tree training, evaluation, and interpretation. section 3.4

















