Gastroenterology Dataset Clustering Using Possibilistic Kohonen Maps

Kohonen maps are an efficient mechanism in signal processing and data mining applications. However, all the existing versions and approaches of this special type of neural networks are still incapable to efficiently handle within a simple, fast, and unified framework, the imperfection of the patterns’ information elements on the one hand like the uncertainty, the missing data, etc., and the heterogeneity of their measuring scale (qualitative, quantitative, ordinal, etc.) on the other hand. Therefore, we propose in this paper a possibilistic Kohonen network essentially based on two fuzzy measures: the possibility and the necessity degrees, to deal with all these aspects together in a robust way. Concrete examples and medical applications will also be given to clarify and to easily explain the proposed algorithm. Key-Words: Self-Organizing Maps (SOMs), Fuzzy Logic (Possibility Theory), Imperfect information, Gastroenterology Dataset, Similarity.