Reducing gaps in quantitative association rules: A genetic programming free-parameter algorithm

The extraction of useful information for decision making is a challenge in many different domains. Association rule mining is one of the most important techniques in this field, discovering relationships of interest among patterns. Despite the mining of association rules being an area of great interest for many researchers, the search for well-grouped continuous values is still a challenge, discovering rules that do not comprise patterns which represent unnecessary ranges of values. Existing algorithms for mining association rules in continuous domains are mainly based on a non-deterministic search, requiring a high number of parameters to be optimised. These parameters hinder the mining process, and the algorithms themselves must be known to those data mining experts that want to use them. We therefore present a grammar guided genetic programming algorithm that does not require as many parameters as other existing approaches and enables the discovery of quantitative association rules comprising small-size gaps. The algorithm is verified over a varied set of data, comparing the results to other association rule mining algorithms from several paradigms. Additionally, some resulting rules from different paradigms are analysed, demonstrating the effectiveness of our model for reducing gaps in numerical features.