@Glen_b , Hi, I am new to SVM, so I have a query that might be a 'trivial' thing, so, instead of asking on Forum, I am asking you.
I designed a feature vector with 26 features to detect face. I programmed using MATLAB and obtained data from 237 face images and 900 background images (all are fixed-size), then fed those data to LIBSVM , and ran various cross validation test (after scaling, using RBF kernel), I got 96.0422% to 98.5928% cross validation accuracy.
Form academic point of view, how 'credible' is this accuracy? Is it good enough? should I go for more data and run cross-validation on that? Thanks.
@Glen_b yes. One was the square of the other. I removed the square, and then the original regressor became significant. I tried searching but didn't find questions related to this. Probably searched with the wrong terms. Could you point me in the right direction? thanks
I thought of multicollinearity, but since one is the square of the other, this doesn't seem to apply in my case.
@Jim, there is no way to answer that question. It depends on the dataset you're working with & the standards of your particular field (eg, the reviewers you'll face). However, in general there is no question too "trivial" for the main site. All statistics / machine learning questions go on the main site, not in chat.
@Anoldmaninthesea., please ask that on the main site, not here. You should search first and read some related threads. You will probably find an answer. At any rate, statistics questions go on the main site; that isn't a chat item.
