日志异常检测

Related Survey:

• Anomaly Detection for Discrete Sequences: A Survey
  • 问题1：已有一个正常的集合，新来一个样本，判断其异常分数
  • Solutions: generative model, non-parametric density estimate, similarity kernel
    • finite state automata (FSA)
    • Cluster method:
      • K-medoid algorithm
      • probabilistic suffux trees
    • Similarity measures:
      • Simple matching coefficient: match two sequences’ similarity
      • Longest common subsequence (LCS), nLCS
    • Markovian Techniques
      • fix markovian techniuqes
      • Variable Markovian Techniques
      • Sparse Markovian Technuques
    • Hidden Markov Model Based Techniques
  • 问题2: 从一个长序列里面检测一个异常的子序列
  • Solutions:
    • window scoring techniques
    • segmentation technique
  • 问题3: 根据序列的频繁程度来判定异常
  • Solutions:
    • Surprise detections in the context of time series data
    • some basic counting method
      • Counting the number ：frequency vs history frequency
      • Counting the number of times a substring of the query pattern’s accurs in a sequence