Standard Decision Trees in Machine Learning - Master’s thesis - Explainability and H2O-3 implementation

Aug 3, 2026

Upload by Miles

Master’s thesis standardizes the Standard Decision Tree (SDT) algorithm and concentrates on practical approaches for creating SDTs with clear construction strategies. The work motivates SDTs as essential instruments for model explainability in Machine Learning by enabling transparent, rule-based reasoning. One construction method is selected and implemented in the H2O-3 Machine Learning platform, followed by testing on a suitable dataset. Results are then compared with an alternative ML model, assessing performance, scalability, and interpretability.

LanguageEnglish

CategoryTechnology

FormatPDF

Pages67

Reading Time2 h 49 min

AccessFree to download

Introduction

Research

Preliminaries

Approaches to Decision Tree creation

ML models explainability with decision trees

Design of Single Decision Tree for H2O-3

Requirements

Implementation notes

Evaluation of DT implementation

Datasets

Comparing the results on data

Scalability

Explainability

Drawbacks and future improvements

Conclusion

Assignment of master’s thesis Title: Standard Decision Trees in Machine Learning Student: Bc. Yuliia Syzon Supervisor: Ing. Adam Valenta Study program: Informatics Branch / specialization: Knowledge Engineering Department: Department of Applied Mathematics Validity: until the end of summer semester 2022/2023 Instructions Describe the Standard Decision Tree (SDT) algorithm. Focus on explaining the approaches to SDTs creation. Explain why SDTs are essential for the explainability of models in Machine Learning (ML). Select one appropriate method to construct SDT and implement it into the H2O-3 Machine Learning Platform. Test your implementation with a suitable dataset and compare it with another alternative ML model. Electronically approved by Ing. Magda Friedjungová, Ph.D. on 21 April 2022 in Prague. Master’s thesis STANDARD DECISION TREES IN MACHINE LEARNING Bc. Yuliia Syzon Faculty of Information Technology Department of Applied Mathematics Supervisor: Ing. Adam Valenta December 27, 2022 Czech Technical University in Prague Faculty of Information Technology © 2023 Bc. Yuliia Syzon. All rights reserved. This thesis is school work as defined by Copyright Act of the Czech Republic. It has been submitted at Czech Technical University in Prague, Faculty of Information Technology. The thesis is protected by the Copyright Act and its usage without author’s permission is prohibited (with exceptions defined by the Copyright Act) . Citation of this thesis: Syzon Yuliia. Standard Decision Trees in Machine Learning. Master’s thesis. Czech Technical University in Prague, Faculty of Information Technology, 2023 . Contents Acknowledgments vi Declaration vii Abstract viii Introduction 1 1 Research 3 1. 1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.1.1 Machine Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.1.2 Decision Trees . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 1.1.3 H2O-3 Machine Learning platform ...................... 13 1.2 Approaches to Decision Tree creation ......................... 16 1.2.1 Standard Decision Tree (SDT) -Greedy approach ............. 18 1.2.2 Oblique decision trees 22 1.2.3 Evolutionary approach-genetic algorithms . . . . . . . . . . . . . . . . . 24 1.2.4 SAT-based Decision Tree Learning . . . . . . . . . . . . . . . . . . . . . . 28 1.2.5 Incremental approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30 1.3 ML models explainability with decision trees . . . . . . . . . . . . . . . . . . . . 32 1.3.1 Explanations via Surrogate Models . . . . . . . . . . . . . . . . . . . . . . 34 1.3.2 Neural-Backed Decision Trees . . . . . . . . . . . . . . . . . . . . . . . . . 36 2 Design of Single Decision Tree for H2O-3 39 2.1 Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 2.1.1 Algorithm (R1 requirement) .......................... 40 2.1.2 Splitting rules (R2-R5 Requirements) .................... 40 2.1.3 Termination rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 2.1.4 Assigning decision values to the leaves (R3, R6 requirements) ....... 41 2.1.5 Optimisations for huge datasets (R7 requirement) ............. 41 2.1.6 Distributed computation (R8 requirement) ................. 42 2.1.7 Presentability (R9 requirement) ........................ 42 2.2 Implementation notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43 3 Evaluation of DT implementation 45 3. 1 Datasets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 3.2 Comparing the results on data ............................ 46 3.3 Scalability ........................................ 47 3.4 Explainability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 3.5 Drawbacks and future improvements 50

FAQs

What is the main focus of the thesis on Standard Decision Trees (SDT)?

The thesis describes the SDT algorithm and emphasizes approaches to SDT creation, including design choices that affect how the trees are built and how they can be used for explanations.

Why are standard decision trees important for explainability in machine learning?

Decision trees support transparent, rule-based decision paths, which helps explain model behavior in a way that is more interpretable than many black-box approaches.

How is the SDT method validated in the H2O-3 platform?

A suitable SDT construction method is implemented in H2O-3, tested with an appropriate dataset, and then compared against another alternative ML model while evaluating results such as scalability and explainability.

Aug 3, 2026

Upload by Miles