Fuzzy Clustering and Mapping of Ordinal Values to Numerical
Mahnhoon Lee, Roelof K. Brouwer · 2007
Classification of object is considered to be the first step in many computationally intelligent systems. Objects are categorized according to their features or characteristics. Objects in the same category can be clustered into groups according to the dissimilarity in terms of their features. These groups reveal some knowledge about the objects by their partitions. Features can be numerical, ordinal or nominal. There has not been a good way to measure the dissimilarity among ordinal values, which is required for clustering. We present a novel algorithm for developing a mapping of ordinal values to numerical values for which a measure of dissimilarity exists. The algorithm is made part of the fuzzy c-means clustering algorithm. The modified algorithm finds better partitioning into clusters as well as an ordinal-numerical mapping that reveals the hidden structural knowledge of the ordinal feature. Simulations show the method to be quite effective