ENTROPY BASED AUTOMATIC CLUSTERING

Abstract

Cluster analysis has been a fundamental research area in data analysis and pattern recognition. In this project, an Entropy based Automatic Clustering method is purposed. It automatically determines the number and initial position of cluster centers. In this, concept of Black hole entropy is used. It calculates the entropy for each data points in a dataset and then select minimum entropy as cluster center. Minimum entropy is chosen because it is the point which is more connected by other data points. Next, it eliminates all those data points which having a criteria greater than threshold value. Again, choose next cluster center in remaining dataset having minimum entropy. This process is repeated until no data points remain in the dataset. Now we will get the appropriate number of centers with their initial location. In order to avoid drawbacks which were occur in EFC, again we check the similarity measure with these obtained cluster centers and put the data point to that cluster for which its similarity value is higher. In this way, we will get better clusters and for making center at mean point, we will calculate the mean of all data points within a cluster and represent it by center of that cluster. There is one parameter which requires to handle i.e. threshold value which is easy to specify. In this method, there is no need to revised entropy value for each data point after cluster center is determined. So, it is simple in nature and also no need of user constraints. Experimental results shows how this method is good for predicting cluster centers. Results are also compared with other clustering algorithms like K-mean and FCM method. Complexity of this method is lesser than standard FCM method. It also handles large dataset very well.