Multi-Objective Feature Selection With Missing Data in Classification

Feature selection (FS) is an important research topic in machine learning. Usually, FS modelled as a bi-objective optimization problem whose objectives are: 1) classification accuracy; 2) number of features. One the main issues real-world applications missing data. Databases with data are likely to be unreliable. Thus, performed on set some also In order directly control this issue plaguing field, we propose study novel modelling FS: include reliability third objective problem. address modified problem, application non-dominated sorting genetic algorithm-III (NSGA-III). We selected six incomplete sets from University California Irvine (UCI) learning repository. used mean imputation method deal experiments, k-nearest neighbors (K-NN) classifier evaluate feature subsets. Experimental results show that proposed three-objective model coupled NSGA-III efficiently addresses for included study.