A-Detector is a software developed to automate the analysis of network anomalies in large dataframes. Thanks to a series of algorithms, A-Detector can detect anomalous data and display it in dynamic graphics.
A-Detector imports network traffic, and based on a series of algorithms like; Variable Scaling and Isolation Forest, is able to normalize data and detect anomalies in the dataframe.
This is the first step to start playing.
import pandas as pd import numpy as np df = pd.read_json('/home/alexfrancow/netflow.json')
At this point we basically tell the application to count the packets that have the same IP and the same protocol in a time period of 5 seconds.
If we have the next table:
ipdst proto time count 10.3.20.102 HTTP 2017-03-20 17:08:56 1 10.3.20.102 HTTP 2017-03-20 17:08:57 1 10.3.20.102 HTTP 2017-03-20 17:08:58 1 10.3.20.102 HTTP 2017-03-20 17:08:58 1 10.3.20.102 TCP 2017-03-20 17:08:59 3
With the data group, the output will be as follows:
ipdst proto time count - - 2017-03-20 17:08:50 0 10.3.20.102 HTTP 2017-03-20 17:08:55 4 10.3.20.102 TCP 2017-03-20 17:08:55 4 - - 2017-03-20 17:09:00 0
An approach to Z-score normalization (or standardization) is the so-called Min-Max scaling. In this approach, the data is scaled to a fixed range - usually 0 to 1.
The IsolationForest ‘isolates’ observations by randomly selecting a feature and then randomly selecting a split value between the maximum and minimum values of the selected feature.
The results of all these algorithms will be printed on a map, to offer a good view to the user.
Tested on Ubuntu 16.04.3 - 18.04
$ sudo apt install python3-venv tshark
$ git clone https://github.com/alexfrancow/A-Detector.git $ cd A-Detector $ . venv/bin/activate $ python run.py
$ git clone https://github.com/alexfrancow/A-Detector.git $ cd A-Detector $ pip install -r requirements.txt $ python3 run.py