Scholarlink

<p>Drug data, particularly toxicity data, poses significant challenges in biomedical research due to its critical role in planning effective treatments. Accurate classification of drugs across various datasets requires robust feature selection (FS) to eliminate irrelevant information and enhance algorithmic performance. This paper presents a novel wrapper FS approach using a hybrid optimization algorithm termed the modified Walrus Optimizer (mWO). This algorithm integrates the strengths of the Rime Optimization Algorithm (RIME) with a local escape operator (LEO) in the Walrus Optimizer (WO) framework, designed to navigate local optima, improve search efficacy, balance exploration and exploitation, and select the most relevant features while preserving classifier accuracy. The rationale for proposing this model is to overcome the key limitations of the original WO: slow convergence, imbalance between exploration and exploitation, and premature stagnation in local optima, while ensuring robust feature selection and improved classification accuracy in high-dimensional chemical and toxicity datasets. The mWO and the original WO were evaluated on the CEC’2022 benchmark and tested on eight diverse datasets (four chemical and four toxicity-related) to validate their performance on complex optimization tasks and FS for real-world datasets. The comparative analysis included binary versions of other established metaheuristics, such as the Whale Optimization Algorithm (WOA), Harris Hawks Optimization (HHO), Grey Wolf Optimization (GWO), Hunger Games Search (HGS), Runge Kutta Optimizer (RUN), and the original WO. Additionally, the mWO was integrated with k-nearest neighbors (k-NN) and support vector machine (SVM) classifiers to enhance chemical data classification. In particular, the mWO-SVM model achieved 98.9% accuracy on the MOA dataset and above 95% on multiple toxicity datasets, clearly outperforming competing metaheuristics. The proposed framework demonstrates potential in advancing drug classification and prediction, offering a versatile computational solution with applications across bioinformatics.</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

Hybrid optimization based feature selection for enhanced chemical data classification using modified walrus optimizer

  • 期刊论文

  • 发表日期：

    2026年3月31日

查看全文
  • Marwa M. Emam,
  • Mosa E. Hosney,
  • Mohammed R. Saad,
  • Nagwan Abdel Samee,
  • Reem Ibrahim Alkanhel,
  • Essam H. Houssein

摘要

Drug data, particularly toxicity data, poses significant challenges in biomedical research due to its critical role in planning effective treatments. Accurate classification of drugs across various datasets requires robust feature selection (FS) to eliminate irrelevant information and enhance algorithmic performance. This paper presents a novel wrapper FS approach using a hybrid optimization algorithm termed the modified Walrus Optimizer (mWO). This algorithm integrates the strengths of the Rime Optimization Algorithm (RIME) with a local escape operator (LEO) in the Walrus Optimizer (WO) framework, designed to navigate local optima, improve search efficacy, balance exploration and exploitation, and select the most relevant features while preserving classifier accuracy. The rationale for proposing this model is to overcome the key limitations of the original WO: slow convergence, imbalance between exploration and exploitation, and premature stagnation in local optima, while ensuring robust feature selection and improved classification accuracy in high-dimensional chemical and toxicity datasets. The mWO and the original WO were evaluated on the CEC’2022 benchmark and tested on eight diverse datasets (four chemical and four toxicity-related) to validate their performance on complex optimization tasks and FS for real-world datasets. The comparative analysis included binary versions of other established metaheuristics, such as the Whale Optimization Algorithm (WOA), Harris Hawks Optimization (HHO), Grey Wolf Optimization (GWO), Hunger Games Search (HGS), Runge Kutta Optimizer (RUN), and the original WO. Additionally, the mWO was integrated with k-nearest neighbors (k-NN) and support vector machine (SVM) classifiers to enhance chemical data classification. In particular, the mWO-SVM model achieved 98.9% accuracy on the MOA dataset and above 95% on multiple toxicity datasets, clearly outperforming competing metaheuristics. The proposed framework demonstrates potential in advancing drug classification and prediction, offering a versatile computational solution with applications across bioinformatics.