I'm dealing with a supervised problem which includes three features. My main purpose is to find the relationship among these features which is complicated. I'm looking for a straight relationship, not a multistage one. Support Vector Regression and Gaussian Process Regression include multiple stages but they are not proper for me. Which algorithm do you recommend to me?
If your three features are your predictors, just make a pearson correlation in order to check if they correlate. One pearson requirement is a normal distributed feature. Avoid multicollinearity ! Then use your predictors in a stepwise regression in order to explain the effect on your Y. SVMs or other ML techniques can be applied for more complicated problems. A simple linear or multiple regression is a good starting point
Seyyed Mohammad Mahdi Bahadornia With only 3 predictor variables, your problem should be easier. If you plot a scatter matrix among three variables, you'll see how are they related. Further, to simplify relationship, try using polynomial regression (quadratic or higher order) should be able to give you the relationship equation.
would someone please explain to me what is PAC learning and empirical risk minimization?
PCA is principle component analysis , this is a method for reducing the dimension of data. By here I am asking about a general concept of machine learning which is the PAC learning : Probably Approximately Correct Learning.
Features and their correlation plays important role i classification accuracy and performance. Refer following paper for dynamic feature selection:
Article Instance based dynamic feature selection for network traffic...
- Badges
- Science topic