东北大学学报:自然科学版 ›› 2020, Vol. 41 ›› Issue (1): 17-22.DOI: 10.12068/j.issn.1005-3026.2020.01.004

• 信息与控制 • 上一篇    下一篇

量子化信息素蚁群优化特征选择算法

李占山1,2, 刘兆赓2, 俞寅2, 鄢文浩2   

  1. (1.吉林大学 计算机科学与技术学院, 吉林 长春130012; 2.吉林大学 软件学院, 吉林 长春130012)
  • 收稿日期:2019-02-01 修回日期:2019-02-01 出版日期:2020-01-15 发布日期:2020-02-01
  • 通讯作者: 李占山
  • 作者简介:李占山(1966-),男,吉林长春人,吉林大学教授,博士生导师.
  • 基金资助:
    国家自然科学基金资助项目(61672261); 吉林省自然科学基金资助项目(2018010143JC); 吉林省发展和改革委员会产业技术研究与开发项目(2019C053-9).

A Quantized Pheromone Ant Colony Optimization Algorithm for Feature Selection

LI Zhan-shan1,2, LIU Zhao-geng2, YU Yin2, YAN Wen-hao2   

  1. 1. College of Computer Science and Technology, Jilin University, Changchun 130012,China; 2. College of Software, Jilin University, Changchun 130012,China.
  • Received:2019-02-01 Revised:2019-02-01 Online:2020-01-15 Published:2020-02-01
  • Contact: YU Yin
  • About author:-
  • Supported by:
    -

摘要: 蚁群优化算法凭借其正反馈机制和强大的搜索能力被广泛地应用于各类优化问题求解上.本文试图将蚁群优化算法应用于特征选择领域并提出了新的量子化信息素蚁群优化(quantized pheromone ant colony optimization, QPACO)特征选择算法.相比于其他基于蚁群优化算法的特征选择算法,QPACO算法中采用了量子化信息素的启发式策略,改变了传统的信息素更新策略,因此避免了在搜索特征时的局部最优问题.实验采用了KNN分类器来指导学习过程,利用源于UCI数据库的多组数据集进行了相关的测试,实验结果表明,QPACO算法在分类精度、精确率、召回率和维度缩减率等方面均具有良好的性能.

关键词: 特征选择, 蚁群优化, 信息素, 量子化, 启发式策略

Abstract: Ant colony optimization algorithms have positive feedback mechanisms and strong searching abilities, which makes them widely used in various kinds of optimization problems. An ant colony optimization algorithm was applied to the field of feature selection and a new quantized pheromone ant colony optimization(QPACO) feature selection algorithm was proposed. Quantum pheromone heuristic strategy was adopted in QPACO algorithm, compared with other ant colony optimization algorithms for feature selection, QPACO algorithm changes the traditional pheromone updating strategy and avoids the local optimization problem when searching for features. In the experimental stage, a KNN classifier was used to guide the learning process, and multiple data sets from the UCI database were used for testing. The experimental results showed that QPACO algorithm has good performances in classification accuracy, precision, recall and feature-reduction.

Key words: feature selection, ant colony optimization, pheromone, quantization, heuristic strategy

中图分类号: