Scholarlink

<p>Machine learning, as a subset of artificial intelligence, is an effective method for data mining and automatic interpretation of issues. Supervised learning plays an increasingly significant role in various medical fields. In this article, a new hybrid classification algorithm that directly combines K-Nearest Neighbors and Naïve Bayes is proposed. This combination aims to leverage the strengths of both algorithms while overcoming their inherent weaknesses to improve accuracy and performance across various tasks, especially on multi-class and imbalanced datasets. The TWV-KNN hybrid algorithm employs a triple-weighting strategy that includes distance and similarity calculations via a kernel function, feature correlations, and posterior probabilities. The performance of the proposed algorithm has been tested on various datasets, with evaluations conducted using K-fold cross-validation, Cohen’s kappa, and Friedman’s test. The evaluation results confirm that the hybrid approach provides a comprehensive view of various types of data, minimizing the likelihood of bias and error in decision-making. Adding Naive Bayes prior probabilities as an algorithmic parameter further highlights the role of the initial conditions defined in the problem in the prediction priorities. Ultimately, Ant Colony Optimization (ACO) is used as an auxiliary feature selection module to identify the most influential attributes in each dataset, thereby enhancing the overall performance of the proposed TWV-KNN classifier.</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

Introducing high performance hybrid multi classification approach for medical data: TWV-KNN algorithm

  • 期刊论文

  • 发表日期：

    2026年1月8日

查看全文
  • Shahrzad Pouramirarsalani,
  • Somayeh Makouei,
  • Karim Abbasian

摘要

Machine learning, as a subset of artificial intelligence, is an effective method for data mining and automatic interpretation of issues. Supervised learning plays an increasingly significant role in various medical fields. In this article, a new hybrid classification algorithm that directly combines K-Nearest Neighbors and Naïve Bayes is proposed. This combination aims to leverage the strengths of both algorithms while overcoming their inherent weaknesses to improve accuracy and performance across various tasks, especially on multi-class and imbalanced datasets. The TWV-KNN hybrid algorithm employs a triple-weighting strategy that includes distance and similarity calculations via a kernel function, feature correlations, and posterior probabilities. The performance of the proposed algorithm has been tested on various datasets, with evaluations conducted using K-fold cross-validation, Cohen’s kappa, and Friedman’s test. The evaluation results confirm that the hybrid approach provides a comprehensive view of various types of data, minimizing the likelihood of bias and error in decision-making. Adding Naive Bayes prior probabilities as an algorithmic parameter further highlights the role of the initial conditions defined in the problem in the prediction priorities. Ultimately, Ant Colony Optimization (ACO) is used as an auxiliary feature selection module to identify the most influential attributes in each dataset, thereby enhancing the overall performance of the proposed TWV-KNN classifier.