Can we use coefficient of determination for nonlinear regression?

3 minute read

Published:

Abstract

This is a note of my thoughts on R^2 after taking Time Series Analysis class by Prof. Byon. I will make the following assumptions.

  1. scalar target: $y \in \mathbb{R}^1$
  2. data is sampled i.i.d.

Introduction

Coefficient of determination arose from the observation in linear regression that

where

The proof is quite ubiquitous in any textbook or online materials. With this equality, one can find a nondimensionalized version of it as, coefficient of determination , $R^2$ as

Note that in the above equation there is two =, which of them is the definition for $R^2​$ is not really certain. Wiki says it is the second one, while Jim claim the first one is more natural. Both of them is well defined and equal in the context of linear regression. In general, we believe the R^2 is a statistic that measures how much proportions of variance of the target is explained by predictor variable, excluding the constant. In such sense, the first one is more natural.

Appearance in nonlinear regression

From my viewpoint, there are mainly two aspects of $R^2$ in the context of linear regression, that makes it popular.

  1. the nondimensional property i.e., we don’t have to worry about getting different performance measure over different datasets. In general, $R^2$ over 0.9 is a good indicator of well performed models. Note that if nondimensional property is not favored, for example, we simply interested in one datasets and we don’t have issue of measuring performance of models across different datasets with different scales, then simply one can choose to use $RMSE$ as commonly seen in machine learning and fluid dynamic community, or the so-called standard error of regression.
  2. variance explanation property. Note that in the context of linear regression, another equivalent phase for explaining variance is correlation coefficient. It can be shown that the square of Person correlation coefficient between $y$ and $\hat{y}$ is essentially $R^2$.. There are other alternative correlation coefficient out there but not really satisfies me anyway. I will write another post about a potentially dumb/workable nonlinear coefficient.

However, as it has been mentioned a lot of times that in the context of nonlinear modeling,

Therefore, one need to make a choice for the definition for $R^2$. Most of the time, people like to use the one with SSE since minimizing SSE is what we do and the smaller the higher for $R^2$. Note that it is implemented in in Scikit-learn as the SSE is well-defined for both linear and nonlinear regression. However, the variance explanation property might not be hold. Because of this issue, there are some negative viewpoint on the usage of $R^2$.

The difference might be small for well-trained nonlinear models

The key to make the equality lies in the following condition

Note that the following two are sufficient conditions for the above,

The second one is easy, as long as model takes constant as linear features

One can show that the OLS solution corresponds to the extrema in $\alpha$ would lead to

For the first one, we notice the following

Note $\epsilon$ is the residual and clearly it is zero mean, if the data is sampled i.i.d, so the explicitly unweighted sum in the above is proportional to the correlation between residual and prediction.

For well-trained, models, Billings et. al., derived several criterions on determining whether the neural network model is well-trained or not. In the equation 21 in their paper, it shows the above uncorrelation in the linear sense is a condition required, which would certainly leads to Under the above sense, we should expect that one is able to use $R^2$ for nonlinear regression with the variance explanation property and nondimensionalized property.