Relational pattern updating

The main goal of this paper is to investigate the problem of updating patterns generated from relational data. For this purpose, we carried out a series of experiments on real databases, showing how relational pattern updating influences the quality of classification. We considered three types of relational patterns: classification rules, classification trees, and class representatives. We propose methods for computing class representatives over relational data and an algorithm that applies such patterns in the process of classification. Another goal is to consider relational data and the patterns generated from this as granules in granular computing. We present methods for the granulation of relational data on different levels, such as elementary granules, sequences of granules, and sets of granules.