Notes

Reviewer #1:
The article deals with an interesting problem. There has been some time I haven't seen an article like this one, well-written and easy to follow. However, I believe the paper does not stress its main contributions.

1. I mean, KddCUP99 dataset is a bit old (I know it's one of the most used one up to date yet), but there are some datasets out there that could be employed as well (NSL-KDD), for instance. That is the first shortcoming this reviewer has observed. Therefore, I believe the authors could address more datasets.

2. In addition, the main contribution is not clear enough. I mean, to address classifiers' combination by means of meta-heuristic-based techniques is not new, i.e., to find out weights considering the voting schema.

The authors could enhance the experimental validation by a cross-validation, instead of using fixed sets.




* In fact we used (NSL-KDD). We only need to clarify any confusion about datasets, and state more clearly what we hoped to acomplish.

*We need to add few paragraphs in introduction and related work sections, which will explain that weights are generated in order to exploit strength of each individual classifier. This is also the novelty of the paper, the related works do not represent a weight model which weighs each classifier's perfomance with each class separately as we do

* Use K-fold cross validation.




Reviewer #3:
1. This paper lacks readability.
(a) The English should be largely improved.
(b) This paper is full of mathematical notations and equations. The authors should try to remove some un-necessary math. notations and equations for improving the readability.
(c) The authors used lots of abbreviations in this paper. Obviously, these abbr. cannot improve the readability. It is the first time for us (readers) that almost all keywords are abbr. in the Keywords section.
(d) The authors should re-organize their paper.
2. Refer to Figure 1: Experimental framework. There are lots of questions and problems in Figure 1. Why the authors selected such combinations using the classfiers using weighted and unweighted majority voting? For an academic paper, the authors should pay more efforts to show the reasons why the adopt/select the (mathematical) combinations!
3. What is the difference between the training data set and the testing data set? Does this (Table 1) make sense?
4. The experimental designs and results are un-convincing. The authors should re-design some experiments in order to demonstrate that their proposed methods can outperform other existing methods in the literature.

In the current manuscript, I suggest rejecting it.

4. the work is realted to spesific dataset "IDS", it is not possible to compare it to some other fields.
2. He is asking to show the mathematical reasons behind that combinations ::::
I think we can do something about that. We can emphasize the importance of diversity of classifiers, and the implemented methods are very diverse

Reviewer #4:
This paper proposes an ensemble classimer configuration for a multi-class intrusion detection problem. The configuration is made up of k-nearest neighbors (kNN), artificial neural networks (ANN), and naive bayes (NB) classifiers. The decisions of these classifiers combined using weighted majority voting (WMV). Two approaches are used to generate effective weight coefficients for WMV: ant colony optimization (ACO) and particle swarm optimization (PSO). I think that is ok. however I have following questions:

1)why do you use Ann,Knn and NB to training dataset, then these classifiers combined using weighted majority voting. in fact you only tell me that you do so, however ,you donot tell me why do you do like this, the reason is more important.many methods, esp some easy methods are combined together, their advantages are ok? I am not sure this result!!

2) your reference are so old, recently 3 years references are so little, I think about these methods, there are many better refereces to discuss this problem.
5)in related work, this part are so easy, only tell me that these methods are useful, however it donot support the necessary and reasonable of your method.


3)intrusion detection system(IDS) are only dataset, in fact, this paper only discuss this method how to realize, not discuss IDS, your paper title is not fit.
4)I think your should analysis IDS, then tell me that what's problem in IDS, about these problems, this method can meet this requriment, hence you use this method to resolve this problem, however, i can not find these content, your creation can not make me believe that this is important. so your creation is not enough.


1. an explanation for the use of ANN, k-NN and NB and weighted majority voting for creating the ensemble.
I think we can do something about that. We can emphasize the importance of diversity of classifiers, and the implemented methods are very diverse