1
$\begingroup$

I was advised to ask my question here.

Recently, I made a post about finding suitable dataset for SIEM (Security Information and Event Management) systems. The goal was to work on classification and correlation to detect security attacks.

I decided to use the dataset from the Honeynet Project Challenges. The problem is this: I don't know if I should use the whole dataset set for my project, because if you look, for example, at the KDD99 dataset, it is devised into two parts: 10% for training and 90% for testing.

I have seen some researcher use dataset A for training and dataset B for testing, do you have any other ideas? I am really stuck at the part of training vs. testing

If my question is too broad. I don't mind some reading materials that will help me deal with my dataset.

Bests,

$\endgroup$
  • 1
    $\begingroup$ Split your data randomly into train and test, the proportion depends on how much data you have and how hard your problem is. $\endgroup$ – user2974951 Oct 22 '18 at 11:27
0
$\begingroup$

You should check out k-fold cross validation techique. Its a quality control technique which makes sure, the prediction/classification done by a model would generalize well to unseen test data. The idea is to split the dataset into k times and use it recursively to train and cross validate your network which at end can be used on the test data to report the actual accuracy.

Here is a link for reference : https://machinelearningmastery.com/k-fold-cross-validation/

$\endgroup$
  • $\begingroup$ Thank you for answer, I will check this link (I might come back later for more clarification if it's possible :) ) $\endgroup$ – U. User Oct 26 '18 at 13:27

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.