Benefits

• Decision paths are intuitive and can be visualised.
• Trees support classification and regression.
• Trees capture non-linear relationships and interactions.

Data and features

• Explore distributions, relationships and outliers before training.
• Use domain knowledge to create meaningful features.
• Limit features when this materially improves interpretability.

Training

• Control max_depth, min_samples_split, min_samples_leaf and max_features.
• Use cross-validation and pruning to limit overfitting.

Evaluation

• Classification: accuracy, precision, recall, F1, ROC AUC and confusion matrix.
• Regression: MSE, MAE and R-squared.

Interpretation

• Visualise the tree and decision paths.
• Review permutation importance, partial dependence and SHAP values.
• Do not treat splits or feature importance as evidence of causality.