A Hybrid K-Mean Clustering Algorithm for Prediction Analysis

Abstract

Background/Objective: The objective of this research is to make improvement in defining the clusters automatically and to assign required clusters to un-clustered points. Methods/Statistical Analysis: The main disadvantage of k-mean is of accuracy, as in k-mean clustering user needs to define number of clusters during the start of process. This restriction of predefined number of clusters leads to some points of the dataset remained un-clustered. So by enhancing the cluster technique, the predictions can be improved. We use Iris dataset for the current study and to generate the results using normalization in the methodology which will lead to improvement in accuracy and will reduce clustering time by the member assigned to the cluster. Findings: The normalization is used to get better results in the form of finding distance to have exact centroid and to remove noise data which is not needed. We are applying backtracking method to find the exact number of clusters that should be defined to analyze the data in better way. The results shows that there is an improvement in clustering when compared to the existing methodologies.