Ali Seyed , Shirkhorshidi (2020) The evolving fuzzy clustering approach for discriminating neutron and gamma-ray pulses / Ali Seyed Shirkhorshidi. PhD thesis, Universiti Malaya.
PDF (Thesis PhD) Download (2969Kb) |
Abstract
Having a significant amount of data is not useful unless the data can be processed for extracting knowledge and information. One of the elementary steps in crunching data is to break it down into groups. When the data is small and collected in a controlled manner, and when the training data is appropriately labelled, the trivial approach is to use supervised learning to perform the grouping. Supervised methods need training data and information about groups beforehand; however, in the current reality, with an avalanche of data, this information is not available. Nevertheless, the need for grouping data remains. Clustering, as an unsupervised method, helps in these situations to group the data. However, unsupervised methods are usually less accurate than their supervised counterparts. To solve this drawback, unsupervised methods are often used as a pre-processing step, along with human judgment, to prune the data to create a reliable training set for the supervised process. One reason that clustering approaches do not yield desirable accuracy is that they will attempt to perform the procedure on all data, which may contain noise or outliers, and they do not have any mechanism by which to set aside the problematic data. Pulse-shape discrimination (PSD) for neutron and gamma-ray pulses that is addressed in this research is one example of a real-world case study that faces the same issues. Although the data utilised for this study is from a liquid scintillator, it can be applied to other signal detectors as well. Aside from this particular dataset, the proposed approach has been applied to a set of publicly available multivariate and time series datasets to prove the performance of the presented approach through an exploratory study. The evolving fuzzy clustering approach (EFCA) proposed in this study utilises a fuzzy membership matrix in fuzzy clustering to propose a new approach for clustering that embeds a heuristic post-pruning solution to address the aforementioned drawback. The method is an EFCA that attempts to find clusters of similar shapes with better accuracy. It introduces an approach for post-pruning that is examined not only on neutron and gamma-ray discrimination but also on various datasets. The outcomes of the proposed method are evaluated against the traditional fuzzy C-means method and another well-known crisp clustering method, namely, K-means. For neutron and gamma-ray discrimination, the EFCA improved the Rand index (RI) accuracy by almost 8%. For other multivariate and time series datasets utilised in this study, results demonstrate the achievement of significant accuracy improvements for some of these datasets after heuristic post-pruning, resulting in 100% RI accuracy for some of them.
Item Type: | Thesis (PhD) |
---|---|
Additional Information: | Thesis (PhD) – Faculty of Computer Science & Information Technology, Universiti Malaya, 2020. |
Uncontrolled Keywords: | Clustering; Fuzzy clustering; Unsupervised learning; Pre-processing; Pruning; Neutron; Gamma-ray discrimination |
Subjects: | Q Science > QA Mathematics > QA75 Electronic computers. Computer science Q Science > QA Mathematics > QA76 Computer software |
Divisions: | Faculty of Computer Science & Information Technology > Dept of Information System |
Depositing User: | Mr Mohd Safri Tahir |
Date Deposited: | 25 Jul 2023 04:10 |
Last Modified: | 25 Jul 2023 04:10 |
URI: | http://studentsrepo.um.edu.my/id/eprint/14652 |
Actions (For repository staff only : Login required)
View Item |