A comparison of fuzzy cluster algorithms in fuzzy clustering analysis

Clustering is an unsupervised learning technique which is used to group samples of data based on their properties of instances. Recently, using of clustering techniques has increased in areas such as bioinformatics, biostatistics, and genetics. Clustering can be performed in two modes, hard and fuzzy. In classical clustering methods, each unit certainly has to be assigned to one cluster. Fuzzy clustering algorithms allow each object to belong to all clusters with a certain membership rating by removing the constraint of assigning to a single cluster. The purpose of this thesis is to provide a detailed transfer of the fuzzy clustering process to the researcher. In such as bioinformatics and genetics areas, it also shows that fuzzy clustering algorithms can be used as an alternative to clustering techniques. In this context, first an implementation was implemented to find the optimal cluster number, which is an important parameter for these techniques. Comprehensive comparisons have been made in the genetic data set commonly used in this application, both using validity indices and elbow method. Then, fuzzy and classical clustering algorithms were applied on the gene expression patterns and compared to measure the performance of the algorithms. Finally, improved fuzzy clustering algorithms have been compared to overcome the problem of outliers, which is a disadvantage of fuzzy clusters. The results of the application have been analyzed in a simple way.