How to get good results when your data has class imbalances.

Using imbalanced-learn to deal with imbalanced data.

Photo by Maria Zavate on Unsplash

In classification problems, we often work with clean roughly balanced datasets in machine learning courses and we don’t have to worry about picking metrics other than accuracy or adjusting our approach to training a model. However, what happens if we end up with a severely class-imbalanced dataset? What happens if we are working with a problem such as fraud detection and only a small fraction of the training samples correspond to instances of fraud?

In this article, I will demonstrate how you can use a Python library called imbalanced-learn to get good results when you are dealing with class imbalances.

Installation

We can install imbalanced-learn with pip using the following command:

pip install -U imbalanced-learn 

We can also install the library with Anaconda as shown below:

conda install -c conda-forge imbalanced-learn

For more information about getting set up with imbalanced-learn, check out the documentation page.

Algorithms for Dealing Imbalanced Data