How to deal with imbalanced datasets?

Sources:

• Dealing with Imbalanced Data (Tara Boyle)

• Having an Imbalanced Dataset? Here Is How You Can Fix I (Will Badr)

Using the correct performance metric

Accuracy is not the best metric to use when evaluating imbalanced datasets as it can be very misleading. It's better to try:

• Confusion Matrix

• Precision

• Recall

• F1

Resampling Techniques

Always split into test and train sets BEFORE trying to resample techniques! And applying resample ONLY in the training set.

Oversample: by adding more copies of the minority class.

Undersample: by removing observations from the majority class.

Create synthetic samples

• SMOTE (Synthetic Minority Oversampling Technique): uses a kNN algorithm to generate new and synthetic data we can use for training our model.

Apply an ensemble algorithm

Ensemble methods