Feature subset selection using multi-objective genetic algorithms

Feature subset selection is a very vast field and it plays a vital role in the modern age because of extremely large datasets with huge number of irrelevant features. In past people have been using several approaches to find subset of features that is most relevant and appropriate. As we explore different techniques we come to know that Genetic Algorithms prove to be exceptionally good in large searches. Even simple genetic algorithms for feature subset selection have produced good results. Recent researcher's show that research is now more focused on Multi-Objective Genetic Algorithms for searching techniques rather than simple genetic algorithms because most of real world examples are multi-objective. Same is the case with feature subset selection, most of the time one subset of features is not of huge interest rather several subsets of features is of interest. So applying multi-objective genetic algorithms rather than simple genetic algorithms produces some great results. This approach is novel and has not been explored to a very large scale. Our research showed that independent sub-sets of features are excellent in accuracy. We have performed this approach on several known datasets present at UCI website for the sake of benchmarking. By observing the results we could conclude that scope of our research does not end here. It could be tried with several variants of multi-objective genetic algorithms as well.