Data mining rules using multi-objective evolutionary algorithms

Cultural Algorithm Toolkit for Interactive Knowledge Discovery

“…Casillas et al, [17] state that the initialization procedure has to guarantee that the initial individuals cover all the input examples from the training data set. To ensure this, the authors of [6] use mutated forms of the default rule as initial solutions where the default rule is the rule in which all limits are maximally spaced and all labels are included. In the current study maximum and minimum chromosomes are used as seeds to create initial population.…”

MO-Miner: A Data Mining Tool Based on Multi-Objective Genetic Algorithms

“…Pareto-based multi-objective GAs have been previously investigated in DM tasks, as the algorithms used in the search for nonlinear models of direct marketing in (Bhattacharyya, 2000) and to select characteristics for non-supervised learning in (Kim et al, 2000). In (Iglesia et al, 2003), an approach based on the NSGAII is employed in the classification task (only one goal attribute), using metrics related to the rules precision: accuracy and covering. A Pareto-based GA was implemented to mine rules from market-basket type databases (association task), using metrics for accuracy, comprehensibility and degree of interestingness (Ghosh & Nath, 2004).…”

Rule Induction for Classification Using Multi-objective Genetic Programming

“…Cover Type datasets, using equal false positive and false negative costs algorithm has been shown to be an effective multi-objective optimizer, both in general and when optimizing rules [1,2,3,4]. Parameter tuning was performed on the Adult dataset and on 10,000 records selected at random from the Cover Type dataset, minimizing the simple error rate and the number of ATs.…”

“…Earlier work by the authors [1,2,3,4] described the application of multi-objective metaheuristics to the problem of partial classification [5]. This problem is the search for simple rules, that represent 'strong' or 'interesting' descriptions of a specified class, or subsets of the specified class, even when that class has few representative cases in the data.…”