Explaining decision tree ensembles with patterns and regressions: in conclusion of a doctoral thesis

It is 2026 and decision tree ensembles are no longer State-of-the-Art on tabular data. What keeps them interesting is that they are based on if-then rules, so they should naturally be symbolic and human understandable. However, the sheer amount of such rules makes the models undeniable black boxes. This talk will present Ordered Pattern Rule Sets: a machine learning model that combines rules, linear regressions, pattern mining, and ordered set theory. We will discuss how it can generalise decision trees, decision lists, rule sets, and their ensembles. Finally, we will show some examples on how to trim decision tree ensembles and how to visually explain both the predictions and the models themselves. The work in its current form has not been published. Yet, it concludes a line of doctoral research started in 2021. We hope that the talk will pave the way to improvements, collaborations, and post-doc contracts in rule-based explainable AI.