Likelihood Based Fuzzy Clustering for Data Sets of Mixed Features

A noble clustering algorithm is presented for data sets of mixed features: numerical, ordinal and nominal. The algorithm uses the concept of fuzzy clustering to reduce negative effect from noises, and uses the iterative partitional algorithm founded on an optimization function to reduce the time complexity. The optimization function uses the likelihood for each individual feature as the optimization criterion of the similarity or likeliness between patterns and clusters, not like the fuzzy c-means clustering algorithm based on distance or the EM clustering algorithm. Hence the algorithm can quickly find fuzzy clusters having different distributions in the each feature level. The simulations show the algorithm to be quite efficient