Switching from Linear to Polynomial Regression
In the field of Machine Learning, Regression is a common term used to define the prediction values of Continuous Dependant Variable.
Depending upon distribution of data, we can determine whether to use linear regression or non-linear regression.
Linear regression is used when the relationship between dependant and independent variable is linear.
When it comes to non-linear data, the primary question which comes in our mind is how can we generate a curve which could capture most part of our data ?
To answer it, polynomial regression comes in handy. It is the simple approach to build non-linear models. It tries to add quadratic terms(square,cube) to a regression.
Why to use Polynomial Regression then ?
When using Linear regression over non-linear data, the outcome is in form of straight line i.e. the straight line doesn’t capture patterns in the data.
This scenario is termed as under-fitting (Highly Biased Model as model results are prone to give more training error).
To give a brief example about it, I have included video which explains polynomial regression using Boston Housing Dataset Example (https://www.kaggle.com/c/boston-housing).
Here is the link to it: https://youtu.be/Vgb9XFa7YyQ
Link to code : https://github.com/Gaurav9112/Polynomial-Regression