Razak, Tajul Rosli
(2020)
Towards designing and measuring interpretable hierarchical fuzzy systems.
PhD thesis, University of Nottingham.
Abstract
This thesis presents a detailed study of the interpretability of hierarchical fuzzy systems (HFSs). It focuses on the development of a design guidelines framework for interpretable HFSs. This thesis aims to fill some of the gaps in the body of knowledge. Several research questions are raised, including: “How can the interpretability of HFSs be measured with indices?”, “How can complexity be comprehensively measured in HFSs?”, “How can user perception on the interpretability and complexity of HFSs be captured?” and “How can interpretable HFSs be designed?” Thus, to study the interpretability of HFSs, this thesis includes the following methodology discussed in different chapters.
First, measures of interpretability are investigated, and an index measuring interpretability specifically in HFSs is introduced. The best way to know about interpretability is by learning how to measure it. Although many researchers have suggested indices to measure interpretability, none of them try to measure the interpretability of HFSs. Indeed, all of them only focus on measuring the interpretability of fuzzy logic systems (FLSs). This is due to the HFSs’ architecture, i.e., multiple subsystems, layers, and topologies, and this presents a significant challenge to measure the interpretability of HFSs. Based on this investigation, this study successfully introduces an initial index for measuring the interpretability of HFSs. The initial index is built based on the challenges arising from the structure of HFSs mentioned.
Due to the subjective nature of the interpretability, the best way to validate the proposed measurements of interpretability of HFSs is by asking the users. However, it is not an easy task to get the user perception, particularly on the interpretability, and there is a lack of research on this issue. Therefore, the second focus of this thesis presents research on a new method of capturing user perceptions of the interpretability and also complexity of HFSs. This is the first time that user study has been used to obtain and assess both qualities in HFSs. Based on this, a new analysis of the relationship between interpretability and complexity of HFSs is presented, and this provides insights into the process of developing measures of interpretability of HFSs.
However, rather than just using the user study to evaluate the measurements directly, this study also uses input from the user study to ‘guide’ the measurement of interpretability of HFSs in what is known as a participatory design approach. The participatory design approach enables the subjective views of a range of users to be taken into account in shaping the measurement of interpretability of HFSs. Thus, the use of the participatory design approach to configure the resulting measurement of interpretability of HFSs is also evaluated.
Complexity is seen as an essential component in determining interpretability. In FLSs, complexity is expressed by the number of rules, variables, and fuzzy terms, called rule-based complexity. Several studies have used indicators (for example, the number of rules) to measure the complexity of FLSs. However, none of the studies considered the structure of HFSs, i.e., multiple subsystems, layers and varied topologies, which may also affect the complexity of HFSs. In addition, the user study revealed different perceptions of complexity in HFSs. Therefore, research is then presented on improving the complexity measurement in HFSs. The measurement is based on combining rule-based complexity with the structural complexity.
Designing an interpretable HFS is a challenging task because of the need to define the interpretability of the architecture of the HFS (the subsystems, the input variables of each subsystem, and the interactions between subsystem), as well as the rules of each subsystem. To assist with this, a design guidelines framework for interpretable HFSs is produced. The framework is based on the measurement index of interpretability and complexity that is presented earlier. The framework consists of five guidelines for building interpretable HFSs.
Finally, to demonstrate a design guidelines framework for interpretable HFSs on the realworld example, a design for an interpretable HFS for a neonatal intensive care unit (NICU) is produced. It is aimed to provide understandable decision support model to clinicians. In the medical context, it is essential for people to understand the importance of the system features. The HFSs is then practically illustrated by using real physiological data at NICU and compared with a flat FLS system. The results show that the design guidelines framework can offer the ability to design an interpretable HFS in practice efficiently.
| Item Type: |
Thesis (University of Nottingham only)
(PhD)
|
| Supervisors: |
Garibaldi, Jon
Wagner, Christian |
| Keywords: |
Hierarchical Fuzzy Systems, Fuzzy Logic Systems, Interpretability |
| Subjects: |
Q Science > QA Mathematics |
| Faculties/Schools: |
UK Campuses > Faculty of Science > School of Computer Science |
| Item ID: |
59844 |
| Depositing User: |
RAZAK, TAJUL
|
| Date Deposited: |
06 Jul 2020 10:32 |
| Last Modified: |
06 Jul 2020 10:45 |
| URI: |
https://eprints.nottingham.ac.uk/id/eprint/59844 |
Actions (Archive Staff Only)
 |
Edit View |
Tools
Tools
![[img]](/style/images/fileicons/application_pdf.png)