Multi-Objective Feature Selection With Missing Data in Classification

Abstract

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