An Efficient K-Means Clustering Initialization Using Optimization Algorithm

Abstract

In data mining has a lot of technique for knowledge discovery. In this Clustering method is very well technique for unsupervised learning. It's important objective is to find a high-quality cluster where the distance between clusters are maximal and the distance in the cluster is minimal. K-means algorithm is applied in this paper for its simplicity. It has been widely discussed and applied in pattern recognition and machine learning. However, the K-means algorithm could not guarantee unique clustering results for the same dataset because its initial cluster centers are select randomly. To avoid such issues a new initialization method is proposed in the Improved K-means algorithm with Cuckoo Search algorithm. The proposed method uses different numerical datasets like iris, wine and solar datasets (Ames, Chariton stations). The K-means clustering solutions are comparable with cuckoo search initialization methods using different measures such as Accuracy, Precision and Recall, F1-score, Silhouette value and MSE (Mean Square Error). The experimental solution represents the effectiveness of the proposed method.