Dear Ali, My research is based on KDD Cup 99 dataset. I have already used it for evaluation and assessment IDS model. Surprisingly i have used Matlab which suit to your question. Please go through my papers to know more about it and how to use KDD Cup 99 in matlab.
After refer the papers , if you have doubt then feel free to contact me.
The short answer is you can't. The underlying dataset itself is far too old and laden with artifacts to be useful for current intrusion detection research. The KDD cup 99 dataset is only a subset of the whole Darpa evaluation subset, so it's even only a part of an already flawed dataset.
There are unfortunately no good alternatives, especially when it comes to labeled datasets. So, make your own and publish it. The research community would be much better served by that than yet another (flawed) study based on the KDD cup 99 dataset. (As a reviewer, we will critizise any study based on this or the underlying DARPA dataset mercilessly).
It really depends what your intentions are. If you are going to use KDD 99 to calibrate an IDS/IPS, then the answer is you should not. On the other hand if you are going to use it to prove a concept then you may do so. For example you would like to investigate whether a neural net can identify malicious traffic by classifying data according to the patterns they produce. Naturally this would be just a concept.
dear Stefan Axelsson, i know it is old. i have used NSLKDD, GUREKDD, Kyoto and OUR PROPOSED dataset NITRIDS . Most of the paper are used KDDCup 99. You cann't say that it is old. Even if it is 16 years old, still researcher used this due to unavailable of public intrusion dataset. KDDCup 99 is best for the beginner in this are.
So i suggest , for beginning purpose , kddcup99 is best.