Data Cube Clustering with Improved DBSCAN based on Fuzzy Logic and Genetic Algorithm

Abstract

Multi-dimensional data, such as data cube, are constructed based on aggregating data in data warehouses and it requires to analyze with high flexibility. Also, clustering, which is an unsupervised pattern recognition analysis, has significant challenges to perform on data cube. In this paper, two new drafts of density-based clustering methods are designed to recognize unsupervised patterns of the data cube. In the first draft, DBSCAN clustering is hybridized by genetic algorithm and called the Improved DBSCAN (IDBSCAN). The motivation of designing the IDBSCAN optimizes the DBSCAN’s parameters by a meta-heuristic algorithm such as GA. The second draft, which is called the Soft Improved DBSCAN (SIDBSCAN), is designed based on fuzzy tuning parameters of the GA in the IDBSCAN. The fuzzy tuning parameters are performed with two fuzzy groups rules of Mamdani (SIDBSCAN-Mamdani) and Sugeno (SIDBSCAN-Sugeno), separately. These ideas are proposed to present efficient and flexible unsupervised analysis for a data cube by utilizing a meta-heuristic algorithm to optimize DBSCAN’s parameters and increasing the efficiency of the idea by applying dynamic tuning parameters of the algorithm. To evaluate the efficiency, the SIDBSCAN-Mamdani and the SIDBSCAN-Sugeno are compared with the IDBSCAN and the DBSCAN. The experimental results, consisted of 20 times running, indicate that the proposed ideas achieved to their targets.