Multiobjective hybrid monarch butterfly optimization for imbalanced disease classification problem

Datasets obtained from the real world are far from balanced, particularly for disease datasets, since such datasets are usually highly skewed having a few minority classes apart from one or more prominent majority classes. In this research, we put forward the novel hybrid architecture to handle imba...

Celý popis

Uloženo v:

Podrobná bibliografie
Vydáno v:	International journal of machine learning and cybernetics Ročník 11; číslo 7; s. 1423 - 1451
Hlavní autoři:	Nalluri, MadhuSudana Rao, Kannan, Krithivasan, Gao, Xiao-Zhi, Roy, Diptendu Sinha
Médium:	Journal Article
Jazyk:	angličtina
Vydáno:	Berlin/Heidelberg Springer Berlin Heidelberg 01.07.2020 Springer Nature B.V
Témata:	Artificial Intelligence Cancer Classification Complex Systems Computational Intelligence Control Datasets Engineering Evolutionary algorithms Genetic algorithms Machine learning Mechatronics Multiple objective analysis Neural networks Optimization Original Article Parameter sensitivity Pattern Recognition Robotics Sensitivity Statistical tests Support vector machines Systems Biology Totally uni-modular matrix Multi-objective optimization SVM Limit-points Evolutionary algorithm
ISSN:	1868-8071, 1868-808X
On-line přístup:	Získat plný text
Tagy:	Přidat tag Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!

Popis
Shrnutí:	Datasets obtained from the real world are far from balanced, particularly for disease datasets, since such datasets are usually highly skewed having a few minority classes apart from one or more prominent majority classes. In this research, we put forward the novel hybrid architecture to handle imbalanced binary disease datasets that arrives upon the efficient combination of Support vector machine (SVM) classifier’s sensitive parameter values for improved performance of SVM by means of an Evolutionary algorithm (EA), namely monarch butterfly optimization (MBO). In this paper, MBO is used to enumerate three objectives, namely prediction accuracy (PAC), sensitivity (SEN), specificity (SPE). Additionally, we propose a Totally uni-modular matrix (TUM) and limit points based non-dominated solutions selection for deciding local and global search and to generate an efficient initial population respectively. Since these two greatly affect the performance of EAs, the performance of the proposed hybrid architecture is tested on 18 disease datasets having binary class labels and the results obtained demonstrate improvements using the proposed method. For the majority of the datasets, either 100% sensitivity and/or specificity were attained. Moreover, pertinent statistical tests were carried out to ascertain the performances obtained.
Bibliografie:	ObjectType-Article-1 SourceType-Scholarly Journals-1 ObjectType-Feature-2 content type line 14
ISSN:	1868-8071 1868-808X
DOI:	10.1007/s13042-019-01047-9