stevenson0421/automatic-feature-selection

self implementation of paper Automated Feature Selection for Anomaly Detection in Network Traffic Data

Automatic-Feature-Selection

self implementation of paper Automated Feature Selection for Anomaly Detection in Network Traffic Data

Experiment Settings

• random seed: 24
• Univariate: mutual information
• selection model: random forest (200 estimators)
• cross validation: 5-fold stratified seperation

Experiment Results

Time Cost

• correlation based
  • NSLKDD 6.4s
  • UNSW_NB15 7.6s
  • IDS2017 145.3s
• Individual
  • Univariate
    • NSLKDD 17.5s
    • UNSW_NB15 9.1s
    • IDS2017 94.0s
  • SFS
    • NSLKDD 5289.2s
    • UNSW_NB15 3002.2s
    • IDS2017 35895.0s
  • RFE
    • NSLKDD 74.0s
    • UNSW_NB15 54.1s
    • IDS2017 564.1s
  • Importance
    • NSLKDD 3.3s
    • UNSW_NB15 2.5s
    • IDS2017 21.4s
• Set
  • NSLKDD 0.003s
  • UNSW_NB15 0.002s
  • IDS2017 0.002s
• Greedy:
  • NSLKDD 6748.8s
  • UNSW_NB15 4220.3s
  • IDS2017 50264.7s

Performance

• NSLKDD
• UNSW_NB15
• IDS2017