计算机工程与应用 ›› 2021, Vol. 57 ›› Issue (5): 47-55.DOI: 10.3778/j.issn.1002-8331.2008-0418

• 热点与综述 • 上一篇    下一篇

新奇检测综述

雷恒林,古兰拜尔·吐尔洪,买日旦·吾守尔,张东梅   

  1. 新疆大学 信息科学与工程学院,乌鲁木齐 830046
  • 出版日期:2021-03-01 发布日期:2021-03-02

Review of Novelty Detection

LEI Henglin, Gulanbaier Tuerhong, Mairidan Wushouer, ZHANG Dongmei   

  1. School of Information Science and Engineering, Xinjiang University, Urumqi 830046, China
  • Online:2021-03-01 Published:2021-03-02

摘要:

能够对异常信息进行检测是智能控制系统的基础能力,新奇检测是一类特殊的异常检测方法,其充分利用了正常数据来构建模型,在诸多智能系统中发挥着重要作用。该领域的综述,能够方便科研人员快速了解新奇检测领域的发展情况,找到适合自己的方法进行应用研究。根据新奇检测方法的基本原理,从基于距离、基于概率、基于域和基于重构四个方面进行了阐述。此外,还介绍了各方法的具体应用以及在经典数据集上的性能表现,并在最后进行了总结分析。研究结果表明,新奇检测方法在工业制造、网络安全、医疗等领域得到了较多应用,具有较好的适应性和广阔的应用前景。

关键词: 新奇检测, K最近邻(KNN)算法, 高斯混合模型, 一类支持向量机(OCSVM)算法, 神经网络

Abstract:

It is a basic ability for intelligent control system to detect abnormal data. Novelty detection is a special kind of anomaly detection, which makes full use of the normal data to build a model, and plays an important role in many intelligent systems. The review of this field can facilitate researchers to quickly understand the recent developments of novelty detection field to find suitable methods for their application research. According to the basic principle of novelty detection method, this paper gives an explication from four aspects:distance-based, probability-based, domain-based and reconstruction-based. Moreover, this paper introduces specific application of each method and the performance on classical datasets, what’s more, it makes a summary analysis in the end. The research results show that the novelty detection methods have been widely used in industrial manufacturing, network security, medical and other fields, showing good adaptability and broad application prospects.

Key words: novelty detection, K-Nearest Neighbor(KNN) algorithm, Gaussian mixture model, One Class Support Vector Machine(OCSVM) algorithm, neural network