Automatic K-Detection Algorithm

Clustering analysis is a important task in data mining. It is a descriptive task that aims to identify groups of similar objects based on the values of their attributes. K-mean algorithm is the most popular partitioning algorithm. As in k-mean algorithm we have to specify the number of clusters in advance. Practically which is very difficult and it also cause performance degradation. With k-mean algorithm we can not find optimal number of clusters. In this paper a new automatic k-detection algorithm(AKD) is proposed which can automatically calculate the number of cluster at run time. This algorithm use the concept of splitting and merging of clusters. For merging the clusters a threshold value is employed while for splitting the clusters standard deviation is used. This algorithm can calculate optimal number of clusters during run time. Experimental results demonstrate that automatic k-detection algorithm can calculate the value of number of clusters (k) automatically and this algorithm also reduces the sum of square error with in cluster. Automatic k-detection algorithm can generate more compact clusters as compare to k-mean algorithm.