I have a dataset of network connections (KYOTO +2006) that each connection has a label of attack or normal. (It doesn't determine which kind of attack).
According to a paper that I want to simulate, I need to choose RANDOMLY and FAIRLY a set of data which contain 1 percent of attack and 99 percent of normal but they did not discuss what does FAIRLY means.
I discriminated data in to attack and normal and randomly choose 1% from attack and 99% from normal but my result was too different from the base paper and I think the problem is in Data selection?
I have a question, Is data selection important in training an IDS (intrusion detection system) and what should I choose training data to be sure it covers all kind of attack and normal data.
Yes, data selection is important in training the IDS since you are trying to identify attacks from your dataset.
So far as choosing of training data is concerned, I would suggest you to initially make protocols for either the clean dataset or the attacking one. Like if you have the list of attacking dataset you can compare all dataset with the existing attacking ones, if matches then block otherwise dataset is clean and so access allowed.
Dear shrikant
thank you for your reply. but can you explain more please
my question is:
I have a data set with label attack or normal, how can I choose a training set from this dataset with 1% attack and 99% of normal to be sure it's representative and it is FAIRE
?
Dear Roya..
I am not sure whether my answer is the one you search for.
From my understanding, you already has labelled dataset (Attack or Normal). You must count the percentage of instances that have been labelled as Attack and Normal. For example if your dataset has 100 instances, and you need 1% attack and 99% of normal; then you should have only 1 instance labelled as attack and the rest are labelled as normal. So these 100 instances is your training dataset.
You should have another set of dataset that will be used for testing. Preferably, testing dataset is not labelled. It can be old dataset or new dataset. The detection results (accuracy) from your testing dataset shows your detection performance/capability.
- Badges
- Science topic
- Similar topics
- Biological Science
- Bioecology
- Systems Ecology
- Landscape Ecology
- Connectivity