Undersampling in Machine Learning Models: Examples & How-to

Undersampling is a strategy used in machine learning to balance datasets that contain a much larger number of one class (the majority) compared to another (the minority). This helps prevent models from ignoring rare but important cases.

1. Example of Undersampling

Credit Card Fraud Detection: Banks use undersampling to balance millions of valid transaction records and a few fraud cases, removing some legitimate transactions so the model can better identify rare fraudulent patterns.​

Spam Email Filtering: Email services reduce the number of “ham” (non-spam) emails in training data, keeping the number similar to “spam” emails. This helps the model focus on spotting rare spam messages despite most emails being legitimate.

2. How Does it Affect Machine Learning Prediction?

Imbalanced datasets can cause machine learning models to focus too much on the majority class, missing the minority class altogether. While undersampling helps by forcing the model to learn to recognize minority cases, removing too many majority examples can also mean losing valuable data. This might reduce the model’s overall predictive power or make it miss patterns present in the majority class.

3. How to Address the Problem of Imbalanced Dataset

• Combine undersampling with other methods, like oversampling (increasing minority samples).
• Test different balancing approaches to see which gives the best model performance.
• Autopilot mode of Aidy can automatically find the data is imbalanced and help to fix the issue.
• Use advanced techniques such as Synthetic Minority Oversampling Technique (SMOTE) or cost-sensitive training.

4. How Aidy Can Help

Aidy simplifies data preparation by automatically detecting imbalance and letting users apply efficient undersampling with just a few clicks. The autopilot platform recommends best practices and visually reports model performance after balancing data, so users without technical backgrounds can still make robust predictions without manual adjustments.

Powered by Google Cloud, Aidy enables rapid and reliable modelling for machine learning projects. You can obtain the free starter pack of Aidy and follow the case study to test it out without any coding requirements and use the function of data feature engineering and data cleanliness. Read more about other common ways to balance the dataset used for modeling.