I'm currently trying to implement a hybrid PSO (Particle Swarm Optimization) to perform feature selection (binary PSO) and parameter optimization (continuous PSO) for a Support Vector Regression machine simultaneously. I have a published article as a reference, but engineering articles usually lack a lot of details.
I'm currently at the 'initialize parameters' phase, so the question is not that complicated. The article said that all input variables were scaled, but later on they use non-scaled values for the SVR parameters. Does anyone know if it will make a difference whether or not to scale the SVR parameters? I'm guessing that if I do scale them, I'll need an extra step to get the original value back for SVR training. However, if I don't scale them, I need 3 different Xmax and Vmax values, so programming-wise it doesn't seem like one is more work than the other.
The article does not seem to be open-source, but for those who have access and want to view the article, it can be found here: http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=5212569
[link] [5 comments]