Intrusion Detection Based on Convolutional Neural Network with Kdd99 Data Set
tensorflow:1.12 python:3.6
Intrusion detection architecture:
An example of data processing for KDDCUP.99 data set:
The basic structure of the convolutional network:
handle5label.py
is the source code to process train data(kddcup.data_10_percent_corrected
) & test data(corrected
).
data set download link:KDD Cup 1999 Data.
cnn_5label.py
is the source code to train CNN.
cnn_test5_label.py
is the source code to test CNN,and count and output each type of classification and fuzzy matrix, in the form as follow:
maybe the matrix or CNN was confused, so i called it confused matrix, not fuzzy matrix in code.
and u need to take some concentrated time to adjust the parameters of the model to solve the problem about imbalance in kdd99 data set. the former result is adjusted by simple oversampling and downsampling. there is a better one in test:
there are something wrong in kdd99 data set,just like the place marked by red circle. http
is true,not icmp
. u can find some other problem by handle5label.py