Classifying malicious network traffic using k-NN

This project uses k-NN to classifiy malcious network traffic in the Intrusion Detection Evaluation Dataset (CICIDS2017) located here .

Specifically, we use the Friday, July 7, 2017 AFTERNOON dataset, which involves a port scan from Attacker Kali 205.174.165.73 on Victim Ubuntu16 205.174.165.68 (Local IP: 192.168.10.50)

The dataset comes with 80 historically useful features already extracted by the CICFlowMeter-V3 .

Feature Selection

From these features, we selected the most dominant using a special technique called Fast Orthogonal Search (FOS). The use of FOS for this purpose will be the topic of a future publication. We have broken the dataset into the following subsets, which are used for training and prediction:

1. training set = 'data/train_inputs.csv'
2. training labels = 'data/train_outputs.csv'
3. test set = 'data/test_inputs.csv'
4. test labels (for evaluation) = 'data/test_outputs.csv'

Note: Malicious traffic was labelled with a 1 and benign traffic was labelled with a -1

Learning

We use the k-NN implementation from the scikit-learn library, which is an ancient technique and well-documented here .

Just run main.py. It should take about 2 mins, depending on your CPU.

Results

Output should look something like this:

• k-NN classified with 99.7 %% accuracy
• The training took 69.5 secs
• The classification took 49.8 secs