Applications of Evolving Tree to Clustering-Based Problems

• Main Author: Chang, Wui Lee
• Format: Thesis
• Language: English; English
• Published: Universiti Malaysia Sarawak (UNIMAS) 2016
• Subjects: T Technology (General)
• Online Access: http://ir.unimas.my/id/eprint/48741/

Clustering is a task of grouping a set of objects in such a way that objects in the same group are more similar to each other than to those in other groups. Examples of popular clustering techniques are K-means, Fuzzy c-Means (FCM), and the Self Organizing Map (SOM). These clustering techniques require a predefined number of clusters. The focus of this thesis is on the Evolving Tree (ETree), one of the relatively new advancements of SOM. ETree is a clustering and visualization technique with incremental learning feature, for constructing a hierarchical (tree) structure, in which the tree structure is allowed to grow to adapt new data objects. ETree is chosen as no predefined number of clusters is needed and the number of clusters increases when new data objects are fed. A search in the literature reveals that the application of ETree to clustering problems is limited. The aim of this thesis is to apply (modified) ETree to two clustering problems, i.e., Failure Mode and Effect Analysis (FMEA) methodology, and textual document clustering. In this first application, ETree is used to cluster and visualize failure modes or corrective actions of FMEA. Such approach is useful for tackling two important shortcomings of FMEA, i.e., the complexity of the FMEA’s worksheet and its intricacy of use. The proposed approach is evaluated with two sets of benchmark and real world information. In this second application, a modified ETree is used to visualize textual documents. The proposed approach is then again evaluated with a banchmark data set and a real case study on UNIMAS flagship Engineering conference (ENCON 2008). The experimental results show that (the modified) ETree is feasible to visualise data objects in a tree structure effectively and improve the learning process (no re-learning is needed). This thesis consitutes of two new applications of ETree to clustering-based problems.