Scholarlink

<p>Feature selection is essential in data mining and machine learning, significantly enhancing classification accuracy and reducing computational overhead. As datasets increase in size, identifying optimal feature subsets becomes more complex, necessitating the use of metaheuristic optimization techniques. Unlike traditional methods, these techniques effectively navigate extensive search spaces. The Snow Ablation Optimizer (SAO), a novel and widely adopted nature-inspired metaheuristic, is recognized for its simplicity and optimization efficacy. Nevertheless, SAO, like other nature-inspired algorithms, encounters limitations such as accuracy constraints, restricted population diversity, and premature convergence, particularly in complex, high-dimensional datasets. To address these limitations, this study proposes an improved variant of the recent optimization algorithm known as the Enhanced Snow Ablation Optimizer (ESAO), along with a wrapper-based feature selection (FS) method that utilizes the k-nearest neighbour (KNN) classifier. In the initial phase, the proposed algorithm employs a position update mechanism using advanced quadratic and Newton interpolation operators to navigate complex search spaces and accelerate convergence. Taylor neighbourhood strategies broaden the search scope, enhancing exploration and solution discovery while minimizing the risk of local optima. Second-order differential perturbation strategies further improve exploration and exploitation by capturing objective function nonlinearities, ensuring more accurate and efficient optimization, and promoting diverse solutions. The SAO algorithm, originally formulated for continuous search spaces, currently employs eight transfer functions: S-shaped, V-shaped, Z-shaped, and three U-shaped, customized to address optimization challenges within binary spaces. Its efficacy was validated using 21 high-dimensional datasets and compared against eight competitor algorithms. Subsequently, the “Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS)” method assesses optimal transfer functions based on classification accuracy, highlighting significant progress with accuracies ranging from 88 to 100%.</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

Enhanced snow ablation optimizer using advanced quadratic and newton interpolation with taylor neighbourhood search and second-order differential perturbation strategies for high-dimensional feature selection

  • 期刊论文

  • 发表日期：

    2025年3月27日

查看全文
  • Shivankur Thapliyal,
  • Narender Kumar

摘要

Feature selection is essential in data mining and machine learning, significantly enhancing classification accuracy and reducing computational overhead. As datasets increase in size, identifying optimal feature subsets becomes more complex, necessitating the use of metaheuristic optimization techniques. Unlike traditional methods, these techniques effectively navigate extensive search spaces. The Snow Ablation Optimizer (SAO), a novel and widely adopted nature-inspired metaheuristic, is recognized for its simplicity and optimization efficacy. Nevertheless, SAO, like other nature-inspired algorithms, encounters limitations such as accuracy constraints, restricted population diversity, and premature convergence, particularly in complex, high-dimensional datasets. To address these limitations, this study proposes an improved variant of the recent optimization algorithm known as the Enhanced Snow Ablation Optimizer (ESAO), along with a wrapper-based feature selection (FS) method that utilizes the k-nearest neighbour (KNN) classifier. In the initial phase, the proposed algorithm employs a position update mechanism using advanced quadratic and Newton interpolation operators to navigate complex search spaces and accelerate convergence. Taylor neighbourhood strategies broaden the search scope, enhancing exploration and solution discovery while minimizing the risk of local optima. Second-order differential perturbation strategies further improve exploration and exploitation by capturing objective function nonlinearities, ensuring more accurate and efficient optimization, and promoting diverse solutions. The SAO algorithm, originally formulated for continuous search spaces, currently employs eight transfer functions: S-shaped, V-shaped, Z-shaped, and three U-shaped, customized to address optimization challenges within binary spaces. Its efficacy was validated using 21 high-dimensional datasets and compared against eight competitor algorithms. Subsequently, the “Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS)” method assesses optimal transfer functions based on classification accuracy, highlighting significant progress with accuracies ranging from 88 to 100%.