An Improvement on Initial Centre Points of K-Means Clustering Algorithm

A large volume of new data is generated every moment of the day by various domains of social media, bank transactions, and online websites. One of the vital means in dealing with this data is to classify it into a set of clusters. At present, many clustering algorithms have been proposed and improved due to the face of large-scale and high-dimensional data, however, they have not achieved better effects and do not generalize well. This paper improves the k-means algorithm, where its results are sensitive to the initial clustering centre, and the clustering results of different similarity measurement methods would be different. The incorporation of in-built polygon sizes with their respective fuzzy linguistic hedge satisfactions for the classification of data points made clusters distinct from each other. To verify the effectiveness of the proposed methods, two sets of data were used as experimental data sets, which produced a good clustering solution with low processing times.