I'm doing a support vector regression with the dependent variable representing measurements from an uncalibrated sensor (measurement error between 2% and 20%) and I want to study the effect of this error on the model performance. What's the best method to inject error on the dependent variable samples ? is it sufficient to do this by adding white noise? Any help or suggestion is appreciated
Adding white noise is simple enough and should work, alternatively you could permute the values of your variable, which is another commonly used method.