11
Network Traffic Classification by Packet Length
Signature Extraction
Srinidhi H, Tamil Esai Somu, Madhusoodhana Chari S
Network Visibility
Dynamic Policy Enforcement
Enhanced Security
Traffic Classification
2
Introduction
Enhanced user
experience!!
Traffic Classification
3
Method
Data Collection
Feature Engineering
Classification
Packet Length Pattern and its Predictability
4
0
500
1000
1500
2000
1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49
VOIP
0
200
400
600
800
1000
1200
1400
1600
1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49
BROWSING
0
500
1000
1500
2000
1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49
FTP forward
FEATURE ENGINEERING
5
Our Solution
 Number of unique packet
lengths
 Statistics of Run Length
Type
Run Length Statistics
No. of
Unique
Packet
Length
Mean Std.
Deviation
Min Max
VOIP 3.26 3.49 1 11 5
FTP 4.75 5.58 1 22 3
Browsing 1.43 1.08 1 7 14
Classification and its Interpretability
6
Tree1: Generated with only packet length
statistics.
Tree2 : Generated with no. of unique packet
length, packet length and run length statistics
Future Expansions
7
 Expanding signature extraction methods –
LZW compression algorithm
 Programming the ASIC
8

IEEE WEICON 2019 Intrepretable Machine Learning using Decision Trees for Network Traffic Classification

  • 1.
    11 Network Traffic Classificationby Packet Length Signature Extraction Srinidhi H, Tamil Esai Somu, Madhusoodhana Chari S
  • 2.
    Network Visibility Dynamic PolicyEnforcement Enhanced Security Traffic Classification 2 Introduction Enhanced user experience!!
  • 3.
  • 4.
    Packet Length Patternand its Predictability 4 0 500 1000 1500 2000 1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 VOIP 0 200 400 600 800 1000 1200 1400 1600 1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 BROWSING 0 500 1000 1500 2000 1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 FTP forward
  • 5.
    FEATURE ENGINEERING 5 Our Solution Number of unique packet lengths  Statistics of Run Length Type Run Length Statistics No. of Unique Packet Length Mean Std. Deviation Min Max VOIP 3.26 3.49 1 11 5 FTP 4.75 5.58 1 22 3 Browsing 1.43 1.08 1 7 14
  • 6.
    Classification and itsInterpretability 6 Tree1: Generated with only packet length statistics. Tree2 : Generated with no. of unique packet length, packet length and run length statistics
  • 7.
    Future Expansions 7  Expandingsignature extraction methods – LZW compression algorithm  Programming the ASIC
  • 8.