AIEOU: Automata-based improved equilibrium optimizer with U-shaped transfer function for feature selection
View/ Open
File version
Accepted Manuscript (AM)
Author(s)
Ahmed, Shameem
Ghosh, Kushal Kanti
Mirjalili, Seyedali
Sarkar, Ram
Griffith University Author(s)
Year published
2021
Metadata
Show full item recordAbstract
The high dimension of any dataset has become an unavoidable challenge in Data Science and MachineLearning. Reducing the number of dimensions by excluding noisy, irrelevant, or correlated information isoften referred to as the feature selection (FS). The ultimate goal in FS is to identify an optimal set ofdimensions (features) of any dataset to develop an efficient learning model, decrease the computational timeand optimize the memory requirement with the help of some methods. Recently, optimization algorithmshave gained popularity in different fields because of their flexibility and ...
View more >The high dimension of any dataset has become an unavoidable challenge in Data Science and MachineLearning. Reducing the number of dimensions by excluding noisy, irrelevant, or correlated information isoften referred to as the feature selection (FS). The ultimate goal in FS is to identify an optimal set ofdimensions (features) of any dataset to develop an efficient learning model, decrease the computational timeand optimize the memory requirement with the help of some methods. Recently, optimization algorithmshave gained popularity in different fields because of their flexibility and effectiveness. Equilibrium optimizer(EO) is a physics-based meta-heuristic algorithm, which is inspired from a well-mixed dynamic mass balanceon a control volume that has good exploration and exploitation capabilities. In this work, an improvedversion of EO is proposed with the inclusion of learning based automata to find proper values of its parametersand AdaptiveβHill Climbing (AβHC) to find a better equilibrium pool. The method is used as a featureselector, evaluated on 18 standard UCI datasets with the help of K-nearest neighbors (KNN) classifier,and compared with eight state-of-the-art methods including classical and hybrid meta-heuristic algorithms.Moreover, the proposed method is applied on high dimensional Microarray datasets which generally containa few samples but a large number of features, and often lead to a ’curse of dimensionality’. The obtainedresults illustrate the supremacy of the proposed method over the other state-of-the-art methods mentioned inthe literature. The source code of this work is available athttps://github.com/ahmed-shameem/Feature_selection.
View less >
View more >The high dimension of any dataset has become an unavoidable challenge in Data Science and MachineLearning. Reducing the number of dimensions by excluding noisy, irrelevant, or correlated information isoften referred to as the feature selection (FS). The ultimate goal in FS is to identify an optimal set ofdimensions (features) of any dataset to develop an efficient learning model, decrease the computational timeand optimize the memory requirement with the help of some methods. Recently, optimization algorithmshave gained popularity in different fields because of their flexibility and effectiveness. Equilibrium optimizer(EO) is a physics-based meta-heuristic algorithm, which is inspired from a well-mixed dynamic mass balanceon a control volume that has good exploration and exploitation capabilities. In this work, an improvedversion of EO is proposed with the inclusion of learning based automata to find proper values of its parametersand AdaptiveβHill Climbing (AβHC) to find a better equilibrium pool. The method is used as a featureselector, evaluated on 18 standard UCI datasets with the help of K-nearest neighbors (KNN) classifier,and compared with eight state-of-the-art methods including classical and hybrid meta-heuristic algorithms.Moreover, the proposed method is applied on high dimensional Microarray datasets which generally containa few samples but a large number of features, and often lead to a ’curse of dimensionality’. The obtainedresults illustrate the supremacy of the proposed method over the other state-of-the-art methods mentioned inthe literature. The source code of this work is available athttps://github.com/ahmed-shameem/Feature_selection.
View less >
Journal Title
Knowledge-Based Systems
Copyright Statement
© 2021 Elsevier. Licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International Licence (http://creativecommons.org/licenses/by-nc-nd/4.0/) which permits unrestricted, non-commercial use, distribution and reproduction in any medium, providing that the work is properly cited.
Note
This publication has been entered as an advanced online version in Griffith Research Online.
Subject
Information and computing sciences
Commerce, management, tourism and services
Psychology
Artificial intelligence
Data management and data science
Machine learning