Selecting Effective Properties in Classification Operations Using Cuckoo Search Algorithm (CSA)

The problem of Feature Subset Selection refers to the concept of identifying and selecting a usefulsubset of features from the initial dataset and is also an important topic in analyzing the degree ofcorrelation in the classification fields that are used to reduce the features set dimensions. This isperformed by eliminating features that make noise or have a low correlation with others. In manydatasets, some features are not involved in decision-making and could be considered as additional. Soselecting a suitable subset of entries can affect both the accuracy of the classification and its speed. Thisstudy proposes a new approach for selecting effective and optimal features using a cuckoo searchalgorithm and two criteria of degree of Mutual information and resolution to calculate the correlationbetween features. Then, by comparing the proposed method with the results of the whole feature set, Frank, and correlation-based feature selection methods, we show that the proposed method is generallyvery efficient. Finally, since the support vector machine has better results for classification, it has alsobeen compared with other feature selection methods in addition to the support vector, which hassuggested better results than the other methods. Since the proposed method introduces a featureselection approach, it can be used in other research areas such as medical diagnostics, damage detectionsystems, or manufacturing systems.

Keywords:

Feature selection , Cuckoo search algorithm , Degree of resolution , Mutual informationwith Category correlation