Six chapters on one of the oldest and most useful ML models. Trees are flowcharts of yes-or-no questions. Combined by the hundred, they remain a strong default for tabular data.
A single tree is easy to read. Most practical tree models combine hundreds or thousands of them, which predicts better and is much harder to read. The chapters build both, with figures where you drag the thresholds and add the trees yourself.
Two moves recur throughout. The first is the split: at each node, pick the question that separates the data best. Trees split on rankings, not distances, so rescaling a feature changes nothing. The second is the ensemble: one tree is unstable and a thousand together are stable, whether they are averaged (bagging, random forests) or added so each corrects the last (boosting). The last two chapters recover some of the readability the ensemble costs, with Shapley values, and then use the same machinery to find where an intervention changes the outcome.