An intelligent technique for pattern-based clustering of continuous-valued datasets

Abstract

In this paper, a novel computationally intelligent technique for Pattern-Based clustering has been proposed. The proposed technique works in two stages. In the first stage feature selection is done. In the second stage a Significance Weight (SW) based Unsupervised Fuzzy Decision Tree Induction (SW-UFDT) is proposed for clustering. The feature selection is done using the ant colony optimization. The feature selection is done using Ant Colony Optimization is done as it is inherently parallel and provides the optimal features. It helps n the dimensionality reduction of the large dataset for clustering. In the second stage an unsupervised decision tree induction method utilizing significance weight of the attribute has been proposed. Finally, the clusters are formed using the equivalence classes. The performance investigation of the proposed technique is done over 20 different capacities UCI repository datasets in terms of two important cluster quality metrics. The comparative analysis shows that the proposed technique is more efficient than the existing Pattern/Non Pattern-Based clustering algorithms. The difference in performance is also validated using the Friedman and Bergmann–Hommel test. Thus, the proposed technique can be used for clustering of large continuous valued data.