- Date: 2018 Dec 17
- Authors: مجید عبدالرزاق نژاد, سیدمجتبی بنائی
- Keywords: Hybrid Attribute Reduction, Independent Component Analysis, Clustering, Particle Swarm Algorithm, Un-supervise Patterns
Development of information technology, design of new data integration methods and improvement of data storage capabilities are causes to produce huge data sets. The dimension of these data are serious challenges for extracting their hidden patterns and knowledge discovery. To solve it, different attribute reduction methods have been introduced either based on component analysis or based on identifying a minimum subset of attributes which can extract the best hidden pattern. In this paper, the combination of these two approaches is considered for the first time. Independent Component Analysis (ICA) as a feature value analyzer and clustering as an unsupervised hidden pattern identifier, combined together to create a new hybrid attribute reduction problem. This problem is solved by the modified version of the particle swarm algorithm. The proposed approach is tested with genetic algorithms, harmonic search and Ant Colony Optimization on the UCI data set and the results show the competitive advantage of this algorithm over other algorithms implemented to solve this problem.
