Introducing a New Classification Method using a CombinedApproach of Machine Learning and Multi-Criteria DecisionMaking

Decision-making issues have become very complicated and it is no longer possible toeasily assume the independence of criteria. Therefore, the use of Analytical HierarchyProcess (AHP) as one of the widely used methods in calculating the weight of criteria,which one of its basic assumptions is non-dependence between criteria, has faced aproblem. Therefore, in order to optimize the parameters of the problem and increasethe classification accuracy, the Particle swarm optimization algorithm will be used.The current research is developmental in terms of its purpose, and quantitative interms of data analysis method and mathematical modeling. In this paper, for the firsttime, a new hierarchical algorithm based on relations between features will bepresented for classification. In fact, in this article, for the first time, by presenting animproved and new version of the particle optimization algorithm, which will haveintersection and mutation operators, the ability to explore and search in the standardoptimization algorithm will be strengthened. Then, by using this new optimizationalgorithm and taking advantage of feature clustering and selecting the final featuresusing the node centrality criterion, a new feature selection method has been presented.The results of comparative studies on credit datasets with different dimensions showedthe very good competitiveness of the proposed method in comparison with knownmachine learning methods. Multi-criteria decision-making methods have often beenused for ranking, while less attention has been paid to the very good ability of thesemethods in data classification. Network analysis process in combination with particleswarm optimization algorithm shows an efficient and appropriate method in the fieldof data classification.

Keywords:

classification , machine learning , multi-criteria decision making , particle swarm optimization