Development of new agglomerative and performance evaluation models for classification

Abstract

This study proposes two new hierarchical clustering methods, namely weighted and neighbourhood to overcome the issues such as getting less accuracy, inability to separate the clusters properly and the grouping of more number of clusters which exist in present hierarchical clustering methods. We have also proposed three new criteria to assess the performance of clustering methods: (1) overall effectiveness which means the product of overall efficiency and accuracy of the clusters which is used to evaluate the performance of the hierarchical clustering methods for the class label datasets, (2) modified structure strength S(c) to overcome the usage problem in hierarchical clustering methods to determine the number of clusters for non-class label datasets and (3) R-value which is the ratio of the determinant of the sum of square and cross product matrix of between-clusters to the determinant of the sum of square and cross product matrix of within-clusters. This will help us to validate the performance of hierarchical clustering methods for non-class label datasets. The evolved algorithms provided high accuracy, ability to separate the clusters properly and the grouping of less number of clusters. The performance of the new algorithms with existing algorithms is compared in terms of newly developed performance criteria. The new algorithms thus performed better than the existing algorithms. The whole exercise is done with the help of twelve class label and six non-class label datasets.