计算机与现代化

• 数据库与数据挖掘 • 上一篇    下一篇

基于稀疏贝叶斯回归的异常检测

  

  1. (南京航空航天大学计算机科学与技术学院,江苏南京210016)
  • 收稿日期:2014-11-13 出版日期:2015-01-19 发布日期:2015-01-21
  • 作者简介:苏乐群(1990-),男,安徽宿松人,南京航空航天大学计算机科学与技术学院硕士研究生,研究方向:模式识别; 冯爱民(1971-),女,江苏南京人,副教授,硕士生导师,博士,研究方向:模式识别,机器学习,系统结构。
  • 基金资助:
    国家自然科学基金资助项目(61170152)

Anomaly Detection Based on Sparse Bayesian Regression

  1. (College of Computer Science and Technology, Nanjing University Aeronautics and Astronautics, Nanjing 210016, China)
  • Received:2014-11-13 Online:2015-01-19 Published:2015-01-21

摘要: 异常检测问题中的数据可以看作是正常信息和异常信息的高度混合,在使得正常信息损失最小的情况下,异常点集合就是前K个包含最多异常信息的样本。启发于这种思想,提出一种基于稀疏贝叶斯回归的异常检测模型,该方法通过在传统的核函数基础上融入Bayesian推理框架,对数据进行回归估计,利用残差法找出偏离程度较大的样本为异常样本。实验结果表明,该方法具有良好的稀疏性和检测精度。

关键词: 稀疏贝叶斯回归, 残差法, 异常检测, 回归估计, 稀疏性

Abstract: The data can be regarded as outliers highly intermixed with normal data in the field of anomaly detection. With minimal loss of normal information in the model, outliers are viewed as the top K samples holding maximal abnormal information in a dataset. Inspired by this idea, an anomaly detection model based on sparse bayesian regression which taken the Bayesian inferring framework into traditional kernel function was proposed to find the sample serious deviated from the model though the result of regress estimation. Experiment results show that this algorithm is of good sparsity and detection accuracy.

Key words: sparse Bayesian regression, residual method, anomaly detection, regress estimation, sparsity

中图分类号: