Two approaches for clustering algorithms with relational-based data

Abstract

It is well known that relational databases still play an important role for many companies around the world. For this reason, the use of data mining methods to discover knowledge in large relational databases has become an interesting research issue. In the context of unsupervised data mining, for instance, the conventional clustering algorithms cannot handle the particularities of the relational databases in an efficient way. There are some clustering algorithms for relational datasets proposed in the literature. However, most of these methods apply complex and/or specific procedures to handle the relational nature of data, or the relational-based methods do not capture the relational nature in an efficient way. Aiming to contribute to this important topic, in this paper, we will present two simple and generic approaches to handle relational-based data for clustering algorithms. One of them treats the relational data through the use of a hierarchical structure, while the second approach applies a weight structure based on relationship and attribute information. In presenting these two approaches, we aim to tackle relational-based dataset in a simple and efficient way, improving the efficiency of corporations that handle relational-based in the unsupervised data mining context. In order to evaluate the effectiveness of the presented approaches, a comparative analysis will be conducted, comparing the proposed approaches with some existing approaches and with a baseline approach. In all analyzed approaches, we will use two well-known types of clustering algorithms (agglomerative hierarchical and K-means). In order to perform this analysis, we will use two internal and one external clusters as validity measures.