As with any statistical method, there are many assumptions that underline Ordinary Least Squares Linear Regression which can easily be overlooked and cause trouble if we don’t check to be sure they are satisfied before implementing the model.
There are 4 key assumptions –
- Linearity meaning that x and the mean of y are linearly related.
- Homoscedasticity of residuals, which means that the residual have a constant variance.
- Normality of residuals. The residuals are the difference between the predicted values and the actual values, and we expect in a linear regression model that these will be normally distributed.
- Finally, that the predictor variables are independent are therefore not correlated with each other (i.e. no multicolinearity ).
If any of these assumptions are violated, then it is not a usable model and should either be tuned or another model type should be used.
Related Posts –
- What happens when OLS Linear Regression Assumptions are Violated?
- Introduction to Linear Regression in Machine Learning