Simultaneous feature selection and clustering with mixed features by multi objective genetic algorithm

Abstract

In this paper, we propose a novel evolutionary clustering algorithm for mixed type data (numerical and categorical). It is doing clustering and feature selection simultaneously. Feature subset selection improves quality of clustering. It also improves understandability and scalability. It unfastens attraction on numerical or categorical dataset only. K-prototype (KP) is a wellknown partitional clustering algorithm for mixed type data. However, this type of algorithm is sensitive to initialization and may converge to local optima. It is optimizing a single measure only i.e. minimizations of intra cluster distance. We have considered clustering as a multi objective optimization problem (MOOP). Minimization of intra cluster distance and maximization of inter cluster distance are two objectives of optimization. Multi objective genetic algorithm (MOGA) is a well-known algorithm which can be applicable for MOOP to find out near global optimal solution. So in this paper we have developed a hybridized genetic clustering algorithm by combining the global search ability of MOGA and local search ability of KP. Experiments on real-life benchmark datasets from UCI machine learning repository prove the superiority of the proposed algorithm.

A Comprehensive Survey on the Process, Methods, Evaluation, and Challenges of Feature Selection

“…2) Bio-inspired methods: Bio-inspired methods attempt to introduce unpredictability into the search process to avoid local optima. Some work on these methods is shown in [226], [227]. 3) Iterative methods: Iterative approaches resolve the UFS issue and reduce the need for a combinatorial search by redefining it as an evaluation problem.…”

A critical review of feature selection techniques for network anomaly detection: Methodologies, challenges, evaluation, and opportunities

“…For instance, in biological and health-care applications [127], socio economics and business [128], software cost predictions [129], and other fields. Most of the current methods (except those proposed in [143] and [144]) have been designed only for numerical data. Consequently, new Unsupervised FS algorithms for mixed data have a lot of scope to be investigated.…”

Laplacian score and genetic algorithm based automatic feature selection for Markov State Models in adaptive sampling based molecular dynamics

“…Feature selection is an unsupervised machine learning task. One method of categorizing them identifies three sub-classes, namely, filters (Devakumari & Thangavel, 2010;Mitra, Murthy & Pal, 2002;Tabakhi & Moradi, 2015), wrappers (Dy & Brodley, 2004;Dutta, Dutta & Sil, 2014;Breaban &Luchian, 2011), andhybrids (Solorio-Fernández, Carrasco-Ochoa &Martínez-Trinidad, 2016;Li, Lu & Wu, 2006). In filter methods, the importance of a feature is studied based on the intrinsic properties of the data.…”