In this paper, three new techniques namely improved Limited Iteration Agglomerative Clustering (iLIAC), Global Outlier Validation (GOV) and Effective Cluster Validation Method (ECVM) are proposed. The proposed work aims to automatically separate the outliers (irrelevant or error data) and normal clusters over the large dataset through the process of identifying the maximum number of highly relative clusters with good accuracy. The first proposed technique iLIAC works with a new threshold (optimum merge cost) that aims to limit the number of iterations, and it automatically identifies the maximum number of highly relative clusters and outliers over the large dataset with higher accuracy and fewer misclassification errors and less computational time. The second technique GOV evaluates the global outliers around the result, and the last technique ECVM measures the purity (intra-cluster similarity) and impurity (intra-cluster dissimilarity) over the result of the iLIAC technique. Experimental results show that the proposed iLIAC technique is quicker and better to separate the normal clusters and outliers over the large dataset with good accuracy than the existing techniques.
Digital Object Identifier (DOI)
R, Krishnamoorthy and Kumar S, Sreedhar
"An Improved Agglomerative Clustering Algorithm for Outlier Detection,"
Applied Mathematics & Information Sciences: Vol. 10:
3, Article 32.
Available at: https://digitalcommons.aaru.edu.jo/amis/vol10/iss3/32