Scholarlink

<p>Cancer microarray datasets often contain many irrelevant, duplicate, and even noisy features, which are likely to reduce the accuracy of classification algorithms. As a branch of feature engineering, the feature selection process aims to improve the classification performance of the desired microarray analysis by restricting the number of features to only those that are specified and valuable. Feature selection is an NP-hard problem, and agents searching for solutions often fall into local optima, requiring increasing time and effort to compute. This implies that a well-designed global search strategy is of utmost importance. Lion optimization (LO) is a recently proposed metaheuristic for global optimization. Due to its biologically inspired pride-based social structure, LO is capable of strong exploration through a nomadic search while ensuring exploitation through cooperative hunting mechanisms. It is well-balanced for optimal feature subset selection in high-dimensional datasets. However, the LO methodology seems to be constructed for continuous optimization tasks. To address this limitation, a variant algorithm, binary LO (BLO), was developed using an S-shaped Transfer Function to address wrapping-based feature selection in microarray cancer datasets. The proposed method was tested on 11 benchmark datasets on cancer microarrays that represent a variety of tumors and high-dimensional feature spaces. In this study, mRMR (Minimum Redundancy Maximum Relevance) is used as a filter method to reduce dimensionality before the wrapper-based BLO optimization method. The efficacy of the mRMR-BLO approach was evaluated across several prominent cancer datasets. It was also compared to four newer binary optimization techniques to test its effectiveness. Various performance metrics, such as Accuracy, MCR, Precision, Recall, Specificity, FNR, FPR, and MCC, are used to evaluate the model. Non parametric Wilcoxon Paired Signed Ranks test is performed to evaluate mRMR-BLO. The results showed that, with smaller feature sets, the proposed mRMR-BLO algorithm achieved the highest prediction accuracy among the compared existing optimization techniques.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Robust feature selection for cancer microarray data using a hybrid mRMR and Binary Lion Optimization Algorithm

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月18日

查看全文
  • Bibhuprasad Sahu,
  • Amrutanshu Panigrahi,
  • Abhilash Pati,
  • B. K. Madhavi,
  • Janmejaya Mishra,
  • Ram Kaji Budhathoki,
  • Saurav Mallik

摘要

Cancer microarray datasets often contain many irrelevant, duplicate, and even noisy features, which are likely to reduce the accuracy of classification algorithms. As a branch of feature engineering, the feature selection process aims to improve the classification performance of the desired microarray analysis by restricting the number of features to only those that are specified and valuable. Feature selection is an NP-hard problem, and agents searching for solutions often fall into local optima, requiring increasing time and effort to compute. This implies that a well-designed global search strategy is of utmost importance. Lion optimization (LO) is a recently proposed metaheuristic for global optimization. Due to its biologically inspired pride-based social structure, LO is capable of strong exploration through a nomadic search while ensuring exploitation through cooperative hunting mechanisms. It is well-balanced for optimal feature subset selection in high-dimensional datasets. However, the LO methodology seems to be constructed for continuous optimization tasks. To address this limitation, a variant algorithm, binary LO (BLO), was developed using an S-shaped Transfer Function to address wrapping-based feature selection in microarray cancer datasets. The proposed method was tested on 11 benchmark datasets on cancer microarrays that represent a variety of tumors and high-dimensional feature spaces. In this study, mRMR (Minimum Redundancy Maximum Relevance) is used as a filter method to reduce dimensionality before the wrapper-based BLO optimization method. The efficacy of the mRMR-BLO approach was evaluated across several prominent cancer datasets. It was also compared to four newer binary optimization techniques to test its effectiveness. Various performance metrics, such as Accuracy, MCR, Precision, Recall, Specificity, FNR, FPR, and MCC, are used to evaluate the model. Non parametric Wilcoxon Paired Signed Ranks test is performed to evaluate mRMR-BLO. The results showed that, with smaller feature sets, the proposed mRMR-BLO algorithm achieved the highest prediction accuracy among the compared existing optimization techniques.