How to read feature importance weights for Linear Regression

Featured image

Let’s get a simplistic view of the output of feature weights for the house pricing dataset. The top 3 most important linear regression features when training to predict a house price are year=2018, location=USA and age.

Categorical Variables

year_2018 is a categorical variable. The weight 0.3145 tells us that the average price of a house is higher by 0.3145 units when the years is 2018 compared to 2017 and 2019 given all other variables are constant.

Continuous Variables

age is a continuous variable. The weight 0.0488 tells us that a unit increase in age (e.g. when the person is one year older) results in an average increase of 0.0488 of the house price.

Weight     Feature
0.3145     year_2018
0.1503     location_USA
0.0488     age
0.0312     year_2017
0.0057     interest
....       ....

Ralted Posts:

This post is a follow up to the Feature Importance with OneHotEncoder and Pipelines in Scikit-learn.

Another related post is the Regression Tips anc Tricks Series I have made.

Reading

I find these articles very useful: