What are the advantages using Principal Components as input for neural networks compared to using "normal input" for neural networks?
Solved – Advantages Using Principal Component Analysis
neural networkspcatime series
Related Solutions
Principal Component Analysis is an unsupervised method -- that is, you use it on independent variables, settings in your case.
Once you have your principal components, you can partition the variance of your independent variables into orthogonal projections. That is, a certain amount of variance goes into component 1 (which is a linear combination of your variables), which is orthogonal to (uncorrelated with, but see here) the other components.
Usually, one uses PCA to replace a large number of variables by a much smaller number of principal components that are (i) uncorrelated and (ii) contain most of the variance.
However, given that you have a large number of samples and only a small number of variables, and given that you have a dependent variable, and given that you want to "explore interaction of settings and their effect on the bit error rate", I don't think PCA is the right method for you. You should consider either linear modelling, linear discriminant analysis, or maybe a method like PLS (which is to PCA what linear regression is to calculation of a mean). Or even a supervised machine learning method (like SVM, random forests, PLS-DA).
The first cutoff, the principal components that explain 50% of the total variance, is indeed suggested based on the authors' experiments on the KDD CUP 99 dataset. Underneath Table 2 they explain that they tested cutoffs between 30% to 70%, and that 50% achieved the highest detection rate at the lowest false alarm rate.
As far as I can tell, they have not mentioned any reasoning behind choosing eigenvalues less than 0.2 but I suspect that they used a similar method. That is, testing various cutoffs between some range, and choosing the cutoff which gives the best results on this dataset.
Slightly unrelated, but very important if you are doing research in Intrusion Detection: be very careful with the DARPA 1998 and KDD CUP 99 datasets. It has been known for a very long time now that these datasets are inherently flawed, and that techniques cannot be accurately evaluated using them [1][2]. The NSL-KDD dataset [2] may be a more reliable evaluation but is still not ideal. Furthermore, there is some interesting debate on the overwhelming use of machine learning and other anomaly detection techniques in intrusion detection research [3]. You might want to read the papers in the reference list for more details.
References:
- McHugh, "Testing Intrusion Detection Systems: A Critique of the 1998 and 1999 DARPA Intrusion Detection System Evaluations as Performed by Lincoln Laboratory"
- Tavallaee et al., "Toward Credible Evaluation of Anomaly-Based Intrusion-Detection Methods"
- Sommer et al, "Outside the Closed World: On Using Machine Learning For Network Intrusion Detection"
Best Answer
You can read a nice description of PCA here.
The advantage of PCA (one method of dimension reduction) in the context of ML (ie. neural networks) would be to reduce the risk of over fitting and reduce computational complexity.
You could also phrase your question as: what are the alternatives to PCA? For PCA vs Partial Least Squares, see this Cross Validated post For PCA vs random subspace projection see this Cross Validated post