Categorical data clustering with evolutionary strategy weighting attributes

Among the clustering algorithms for categorical data, the fuzzy k-modes algorithm is an effective one. However, it considers that the attributes of data have the same influence on the clustering result. An improved clustering algorithm is presented, assuming the different contribution of attributes of data to the clustering and giving each of them a weight. With a new fitness defined, the evolutionary strategy is used to optimize the weighting matrix of attributes. The clustering accuracy based on the partition similarity is used to evaluate the clustering result. With the little soybean disease data set as the input, the experiment indicates an improved result. Moreover, the weights optimized can be used to extract attributes and reduce the dimensions of data.