ABSTRACT Transfer functions have a very important role in metaheuristic optimization‐based feature selection algorithms as these functions map the continuous search space into binary space. The U‐shaped transfer function (UTF) is one of the transfer functions used to solve the problem of feature selection. However, the UTF requires the selection of parametric values, which can vary for different types of data. To address this issue, an approach to select the parameters of the UTF has been proposed based on a time‐varying adaption method, resulting in the modified U‐shaped transfer function (MUTF). Furthermore, a methodology has been proposed to enhance feature selection and classification for Parkinson's disease by utilizing z‐score normalization in conjunction with a modified U‐shaped transfer function and the binary self‐adaptive bald eagle search (MUTF‐SABES) optimization algorithm. The z‐score normalization has been used to mitigate issues caused by outliers. Also, the performance of the k nearest neighbor classifier is improved by selecting an optimal parameter value using the proposed MUTF‐SABES algorithm. The effectiveness of the proposed methodology is validated on seven different Parkinson's disease datasets and compared with five state‐of‐the‐art optimization algorithms: Salp Swarm algorithm, Harris Hawks optimization, equilibrium optimizer, aquilla optimizer, and Honey Badger algorithm, to evaluate its performance superiority. The results achieved using the proposed approach have been superior or analogous to the erstwhile algorithms for performance comparability. Friedman's mean rank test is used to check the statistical significance of the propounded approach. The lowest Friedman's mean rank value obtained using the proposed approach indicates that the proposed approach has the potential to become an alternative to other well‐known strategies.

Summary Metaheuristic techniques have gained the attention of many researchers in the last few years. These techniques are used to solve real‐world optimization problems as well as for feature selection. This work proposes a lévy flight and disrupt operator‐based elephant herding optimization algorithm (LDEHO). In the proposed algorithm, opposition based‐learning is utilized to commence with better initial positions. Lévy flight and matriarch mean are introduced to update the positions of clan individuals. In addition, the disruption phenomenon is introduced to generate new individuals for replacing the worst clan individuals. Further, an elitism scheme is introduced to preserve the best search agents in consecutive iterations. The performance of the proposed LDEHO algorithm is validated on 97 benchmark functions. A comparative analysis of the proposed LDEHO algorithm with fourteen state‐of‐the‐art algorithms has been made. Results show the high potential of the LDEHO algorithm in solving the benchmark functions. Further, Friedman's mean rank test and multiple comparison tests are applied to demonstrate the statistically significant difference between the algorithms. Moreover, a binary version of the proposed LDEHO algorithm is introduced for feature selection to classify the medical datasets. The performance of the binary LDEHO is validated on 15 medical datasets and compared with six state‐of‐the‐art algorithms. Results show the supremacy of the binary LDEHO for feature selection to classify the medical data. Friedman's test proves that the proposed binary LDEHO algorithm is statistically different and better than other algorithms.

Abstract According to the world health organization report, brain cancer has the highest death rate. magnetic resonance imaging (MRI) for detecting brain tumours is adopted these days due to several advantages over other detection techniques. This paper presents a novel methodology to classify MR images based on texture and deep features, z‐score normalization, and, Comprehensive learning elephant herding optimization (CLEHO) based feature optimization and classification. Deep features of brain MR images have been extracted through DenseNet121 convolutional neural network and texture features have been extracted by using the Gabor 2D filter, Haralick texture feature, edge continuity texture feature, first order statistical texture feature, local binary pattern feature, difference theoretic texture feature, and spectral texture feature techniques. Normalization has been done using three normalization techniques that is, z score, mean median absolute deviation (MMAD), and Tanh‐based after aggregating the features extracted from the previous step. z‐score normalization has been suggested for feature normalization after comparing the results attained from the three techniques. Lastly, binary CLEHO has been proposed for selecting an optimal feature set and also optimizing the ‘k’ value of the k‐NN classifier. The outcome of this proposed work is compared with other state‐of‐the‐art methods for a publicly available magnetic resonance image Fighshare dataset of 3064 slices from 233 patients. The proposed work has a brain tumour average classification accuracy of 98.97%, which is better than the other state‐of‐the‐art methods. The proposed work can be used to assist the radiologist in the screening of multi‐class brain tumours.

Abstract Bald Eagle Search optimization (BES) is introduced recently, which mimics the bald eagles' hunting and food searching behaviour. The capability of a BES algorithm is enhanced in this paper by avoiding local optima stagnation and premature convergence problems. The BES algorithm is modified to enhance the performance of the algorithm and the modified algorithm is called the Self Adaptive Bald Eagle Search (SABES) algorithm. Dynamic‐opposite Learning (DOL) method is invoked in the initialization phase to increase the population diversity and convergence speed. To find better global solutions, the exploitation capability of the SBES algorithm is enhanced by considering the dynamic‐opposite solutions. In addition, the algorithmic parameter values of the BES algorithm have been determined using the linear and non‐linear time‐varying adaption strategy to create a balance between the search abilities which promotes the overall performance of the algorithm. The performance of the SABES algorithm is validated by comparing the results of 50 benchmark functions and CEC2017 functions with different erstwhile algorithms. The proposed algorithm achieves the best results in 80% of the benchmark functions, whereas the BES only gets the best results in 56% of functions. For the CEC2017, the SABES algorithm achieves optimal results for 20 functions which are highest in comparison to state‐of‐the‐art algorithms. The feasibility and effectiveness of the proposed algorithm are checked using the 15 CEC2020 competition real‐world single‐objective constrained optimization problems. The SABES achieves a 100% success rate and 100% feasibility rate in comparison to other well‐regarded algorithms. The statistical significance of the algorithm has been proved using Friedman's mean rank and Wilcoxon sign rank test.

Abstract In this article, a hybrid algorithm has been proposed for the identification of phishing and legitimate websites. The dataset may have an imbalanced class distribution and may consist of irrelevant features. Therefore, in the data preprocessing, the adaptive synthetic sampling approach has been used to handle the imbalanced data. Irrelevant or redundant features are removed from the balanced data using the proposed binary version of Rao algorithms. The S‐shaped and V‐shaped transfer functions are applied for mapping continuous search space to discrete search space. Also, the results of these S‐shaped and V‐shaped transfer functions are analyzed for proposed algorithms. The performance is improved by optimizing the value of the k parameter in the kNN classifier. The dataset used in this article has been taken from the UCI machine‐learning repository. The performance of the proposed approach has been evaluated using the polygon area metric. The obtained classification accuracy is 97.044%. A comparison of the proposed hybrid algorithm with the other state‐of‐the‐art techniques is also made for validation. Moreover, the proposed approach has been compared with seven metaheuristic feature selection algorithms and six filter methods for performance analysis. Additionally, we have applied the proposed approach to URLs that are registered on the PhishTank website.

Abstract Machine learning algorithms are widely applied to biomedical data to classify the samples of patients and healthy persons. The high‐dimensional biomedical datasets contain a large number of features to represent a sample. However, such datasets may have redundant, noisy and irrelevant features, influencing machine learning algorithms' classification performance and increasing computation overhead. Therefore, data normalization and feature selection techniques are introduced to reduce the impact of noisy features and accurately identify the patterns in features to improve the predictive accuracy of diagnosis. This work proposes an efficient feature selection and parameter optimization method to classify high‐dimensional biomedical datasets. In the proposed method, a binary version of the improved elephant herding optimization (IEHO) algorithm is introduced to select features and optimize the C and γ parameters of the support vector machine classifier. Further, four variants of the proposed method are presented based on data normalization techniques: Z ‐score normalization (ZN), Pareto‐scaling (PS), tan h ‐based normalization (TN), and variant of tan h ‐based normalization (VTN). The proposed variants reduce the dominance of noisy features and explore the feature space to obtain the optimal feature set that maximizes the classification accuracy and minimizes the time complexity. The performance of the proposed variants is evaluated on 15 high‐dimensional biomedical datasets. Friedman's mean rank test is applied to check the statistical difference between proposed variants. Results show that the proposed Z ‐score normalization‐IEHO (ZN‐IEHO) variant performed significantly better than the other proposed variants for classification accuracy, false‐positive rate and f ‐score metrics. Moreover, the performance of the proposed ZN‐IEHO variant is compared with 18 state‐of‐the‐art feature selection methods. The experimental results expressed the effectiveness of the proposed ZN‐IEHO variant in finding the best combination of features and parameters to classify the biomedical datasets accurately.

Abstract Rao algorithms are recently proposed optimization algorithms used to solve optimization problems. These algorithms are based on the best and the worst solutions, which are computed during the optimization process. However, these algorithms apply to continuous problems only. In this article, the binary versions of Rao algorithms are proposed, which can be used for solving feature selection problems. These are applied to four publicly available Parkinson's disease datasets. Besides providing an optimal set of features, the k parameter of the k‐nearest neighbour classifier is also optimized by the proposed approach. The performance of these algorithms has been measured taking an average of 30 independent runs using a 10‐fold cross‐validation procedure. Also, a comparison of the performance has been made with the other state of the art methods. Significance analysis of these algorithms has been made with the Friedman rank test.