PhD thesis "Unification of Decision Support Techniques: Mitigating Statistical Paradoxes for Enabling Trustworthy Decision Making" is all about improving decision-making tools that are used in various fields. The author has created a framework to make these tools more reliable and effective and discussed various ways to handle statistical paradoxes in machine learning.

This doctoral thesis provides six significant contributions to address existing technological and knowledge gaps to foster fair and trustworthy decision-making processes. These contributions are the outcomes of two primary research questions and six supplementary research questions answered within the thesis. The thesis utilizes design science research methodology to create innovative artifacts and methods, providing new insights to widen understanding of the domain under the research.

The first contribution provides ways to establish semantic correspondence between the three major decision support techniques, i.e., statistical reasoning, online analytical processing and association rule mining. It examines various approaches to bridge the gap between them. The second contribution is a novel framework for unifying decision-support techniques for developing a unified platform to interpret results from one DST (Desicion Support Technique) to another. The third contribution discusses two measures for identifying confounding effects in categorical and continuous datasets. The fourth contribution discusses the measure for adjusting the confounding effects. Further, the fifth contribution provides a framework for mitigating the impact of bias resulting from statistical paradoxes. The sixth contribution is a web-based application that automatically detects and addresses confounding effects. This application is an invaluable tool for data scientists and researchers, offering automated detection and mitigation of confounding effects and providing a streamlined approach to effectively addressing and overcoming such data analysis challenges. The author argues that the suggested framework and application possess substantial potential for further extensions beyond their current scope of application.

The thesis is published in the Digital Collection of TalTech Library.

Supervisor: Prof Dirk Draheim.

Opponents:

• Divesh Srivastava, PhD, Database Research Department, AT&T Labs, New Jersey, USA;
• prof Ladjel Bellatreche, Laboratoire d’Informatique et d’Automatique pour les Systèmes École Nationale Supérieure de Mécanique et d’Aérotechnique Poitiers, France.

Follow public defence in Zoom

Meeting ID: 997 4921 7667
Passcode: 102947