Section Recap
Introduction
Congratulations, you've just modeled some complex relationships with interaction terms, polynomials, and regularization! Here is a recap of what you learned in this section.
Objectives
You will be able to:
- Critically think about the different ways data scientists add complexity to basic regression models.
- Determine the right balance between complexity and simplicity through the lens of bias and variance
Key Takeaways
This section gave you the chance to learn about techniques whereby you can model non-linear relationships with data. It's very rare that real-world problems can be modeled with a simple linear regression, so it's important to get yourself well acquainted with creating new features and selecting the most important ones.
- An interaction is a particular property of three or more variables, where two or more variables interact in a non-additive manner when affecting a third variable.
- Polynomial regression allows for better fitting data that isn't well predicted using a linear model.
- The risk of polynomial regressions is that it's easier to overfit data, so it's important to consider the Bias-Variance trade-off and perform proper cross-validation.
- Ridge and Lasso regressions are two regularization techniques used for making complex models more expensive in the cost function, reducing the risk of overfitting.
- Feature Selection is an important component of model building, and it can have a drastic impact on the overall performance of a model.
- AIC and BIC are techniques for selecting models that penalize models the more complex they are.
Summary
Excellent work! You learned a substantial amount about different ways to model non-linear relationships. You will continue to use and build upon the concepts learned in this section for the rest of your machine learning career.