Abstract

Feature selection is one of the key components of data mining and machine learning domain that selects the best subset of features with respect to target data by removing irrelevant data. However, it is a complex task to select optimal set of features from a dataset using traditional feature selection methods, as for n number of features, $$2^n$$ feature subsets are possible. Therefore, this paper introduces a novel metaheuristics-based feature selection method based binomial cuckoo search. Generally, metaheuristics-based feature selection methods suffer with stability issue since they select different set of features in different runs. Hence, to deal with stability issue, a hybrid data transformation method based on principal component analysis and fast independent component analysis has also been introduced. The proposed hybrid data transformation method first transforms the original data thereafter proposed binary binomial cuckoo search method is used to elect the best subset of features. The proposed feature selection method maximizes the classification accuracy and minimizes the number of selected features. The performance of the proposed method has been tested on the fourteen feature selection benchmark datasets taken from UCI repository and compared with other latest state-of- the art approaches including binary cuckoo search, binary bat algorithm, binary gravitational search algorithm, binary whale optimization with simulated annealing, and binary grey wolf optimization. Further, statistical analysis has also been carried out to validate the efficacy of the proposed method.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call