Fuzzy Criteria in Multi-objective Feature Selection for Unsupervised Learning

Feature selection in which most informative variables are selected for model generation is an important step in pattern recognition. Here, one often tries to optimize multiple criteria such as discriminating power of the descriptor, performance of model and cardinality of a subset. In this paper we propose a fuzzy criterion in multi-objective unsupervised feature selection by applying the hybridized filter-wrapper approach (FC-MOFS). These formulations allow for an efficient way to pick features from a pool and to avoid misunderstanding of overlapping features via crisp clustered learning in a conventional multi-objective optimization procedure. Moreover, the optimization problem is solved by using non-dominated sorting genetic algorithm, type two (NSGA-II). The performance of the proposed approach is then examined on six benchmark datasets from multiple disciplines and different numbers of features. Systematic comparisons of the proposed method and representative non-fuzzified approaches are illustrated in this work. The experimental studies show a superior performance of the proposed approach in terms of accuracy and feasibility.