A new initialization method for categorical data clustering

In clustering algorithms, choosing a subset of representative examples is very important in data set. Such ''exemplars " can be found by randomly choosing an initial subset of data objects and then iteratively refining it, but this works well only if that initial choice is close to a good solution. In this paper, based on the frequency of attribute values, the average density of an object is defined. Furthermore, a novel ini-tialization method for categorical data is proposed, in which the distance between objects and the density of the object is considered. We also apply the proposed initialization method to k-modes algorithm and fuzzy k-modes algorithm. Experimental results illustrate that the proposed initialization method is superior to random initialization method and can be applied to large data sets for its linear time complexity with respect to the number of data objects. Clustering data based on a measure of similarity is a critical step in scientific data analysis and in engineering systems. A common method is to use data to learn a set of centers such that the sum of squared errors between objects and their nearest centers is small (Brendan & Delbert, 2007). At present, the popular partition clustering technique usually begins with an initial set of randomly selected exemplars and iteratively refines this set so as to decrease the sum of squared errors. Due to the simpleness, random initiali-zation method has been widely used. However, these clustering algorithms need to be rerun many times with different initializa-tions in an attempt to find a good solution. Furthermore, random initialization method works well only when the number of clusters is small and chances are good that at least one random initializa-tion is close to a good solution. Therefore, how to choose initial cluster centers is extremely important as they have a direct impact on the formation of final clusters. Based on the difference in data type, selection of initial cluster centers mainly can be classified into numeric data and categorical data. Aiming at numeric data, several attempts have been reported to solve the cluster initialization to date, few researches are concerned with initialization of categorical data. Huang (1998) introduced two initial mode selection methods for k-modes algorithm. The first method selects the first k distinct objects from the data set as the initial k-modes. The second method assigns the most frequent categories equally to the initial k-modes. Though the second method …