Mining Classification Rules Using Evolutionary Multi-objective Algorithms.

Evolutionary-based methods provide a framework for mining classification rules, that is, rules that can be used to discriminate between data organized in several classes. In this paper, we propose a novel multi-objective extension for the standard Pittsburg approach. Key features of our model include (a) variable length chromosomes, implemented using an active bit string (mask), and (b) fitness evaluation and selection based on restricted non-dominated tournaments. Extensive numerical simulations show that the proposed algorithm is competitive with - and indeed outperforms in some cases - other well-known machine learning tools using benchmark datasets. [ABSTRACT FROM AUTHOR]