A Novel on Altered K-Means Algorithm for Clustering Cost Decrease of Non-labeling Big-Data

Machine learning in Big Data is getting the spotlight to retrieve useful knowledge inherent in multi-dimensional information and discover new inherent knowledge in the fields related to the storage and retrieval of massive multi-dimensional information that is newly produced. The machine learning technique can be divided into supervised and unsupervised learning according to whether there is data labeling or not. Unsupervised learning, which is a technique to classify and analyze data with no labeling, is utilized in various ways in the analysis of multi-dimensional Big Data. The present study thus proposed an altered K-means algorithm to analyze the problems with the old one and determine the number of clusters automatically. The study also proposed an approach of optimizing the number of clusters through principal component analysis, a pre-processing process, with the input data for clustering. The performance evaluation results confirm that the CVI of the proposed algorithm was superior to that of the old K-means algorithm in accuracy.