Multiple Linear Regression RMSE vs. number of features/predictors

1 次查看（过去 30 天）

Aalaas 2015-10-27

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/251322-multiple-linear-regression-rmse-vs-number-of-features-predictors

Hello everybody,

I am trying to evaluate the performance of my MLR regression model in terms of the number of input features (or predictors). I have first computed the p-values of each predictor for the result, and then, I have sorted the predictors in such a way that their p-values are in an increasing order. Later, I created N regression models (where N=number of predictors), each one of them adding a new predictor to the model. I computed the RMSE results for the models, and ploted them like: number of predictors used vs. RMSE for each model. I hoped to have a nice curve with a decreasing RMSE as the number of features increased, but not at all. In fact, the RMSE seems to be totally random.

Is there a simple way to obtain such a curve? Or is it necessary to use algorithms such as the Sequential Feature Seection for this purpose?

All your ideas are welcome! Thank you very much!