A Two-Step Iterative Procedure for Clustering of Binary Sequences

Association Rules (AR) are a well known data mining tool aiming to detect patterns of association in data bases. The major drawback to knowledge extraction through AR mining is the huge number of rules produced when dealing with large amounts of data. Several proposals in the literature tackle this problem with different approaches. In this framework, the general aim of the present proposal is to identify patterns of association in large binary data. We propose an iterative procedure combining clustering and dimensionality reduction techniques: each iteration involves a quantification of the starting binary attributes and an agglomerative algorithm on the obtained quantitative variables. The objective is to find a quantification that emphasizes the presence of groups of co-occurring attributes in data.