NICGAR: A Niching Genetic Algorithm to mine a diverse set of interesting quantitative association rules

• We present NICGAR, a Niching Genetic Algorithm to mine a set of different positive and negative quantitative association rules.
• This proposal includes two threshold values that allow the user to adjust the diversity and quality of the obtained rules.
• We propose distOC, a new distance measure between rules based on the examples covered and the common attributes of the rules.

Evolutionary algorithms are normally applied to mine association rules on quantitative data but most of them obtain enough similar rules due to that the usual behavior of these algorithms is to converge on the best solution of the problem. To overthrow this issue, in this paper we present NICGAR, a new Niching Genetic Algorithm to obtain a reduce set of different positive and negative quantitative association rules with a low runtime. To do that, we extract the rules based on the existence of a pool of external solutions that contains the best rule of each niche found in the search process according to several quality measures, we penalize similar rules by means of a process based on fitness sharing, and we restart the algorithm leading to a diverse population. Moreover, the user can tune the trade-off between the quality and diversity of the mined rules making use of two thresholds. Finally, a new measure has also been presented to assess the similarity between rules based on involved attributes and covered examples by the rules. The quality of our proposal is analyzed using statistical analysis and comparing with classical, mono-objective evolutionary, and multi-objective evolutionary approaches for mining association rules.