Robust OCTs

Investigating classification tree robustness

Bachelor Thesis (2022)
Author(s)

G. Lek (TU Delft - Electrical Engineering, Mathematics and Computer Science)

Contributor(s)

Krzysztof Postek – Mentor (TU Delft - Discrete Mathematics and Optimization)

Faculty
Electrical Engineering, Mathematics and Computer Science
Copyright
© 2022 Gert Lek
More Info
expand_more
Publication Year
2022
Language
English
Copyright
© 2022 Gert Lek
Graduation Date
20-06-2022
Awarding Institution
Delft University of Technology
Programme
['Applied Mathematics']
Faculty
Electrical Engineering, Mathematics and Computer Science
Reuse Rights

Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.

Abstract

The application of machine learning in daily life requires interpretability and robustness. In this paper we try to make the process of building robust and interpretable decision trees more accessible. We do this by making the fitting of these models cheaper and simpler. We build on previous research and see if changing input data or the fitting formulation can create more robust trees that can be computed faster. To investigate this, we test whether data perturbations make heuristic algorithms more robust and whether enforcing constraints on adversarial examples in normal optimal classifica- tion tree MILP formulations can improve robustness. We also provide an altered formulation for the robust OCT model in Vos and Verwer (2021b) that yields better results with shorter runtimes. Finally, we extend the ROCT formulation to be applicable to multi-class classification and regression tasks.

Files

License info not available