Soledad Galli, PhD

@solegalli.bsky.social

110 Followers 23 Following

Python Open-Source Developer | Developer Advocate | Scientific Python & Machine Learning | AI Educator | Author & Speaker Find out more at Train in Data: https://www.trainindata.com/

3D hardcover mockup of Imbalanced Data: Myths, Mistakes and Modern Solutions
Book cover mockup for Imbalanced Data: Myths, Mistakes and Modern Solutions displayed on a tablet
Book cover mockup for Imbalanced Data: Myths, Mistakes and Modern Solutions displayed on a tablet
3D hardcover mockup of Imbalanced Data: Myths, Mistakes and Modern Solutions
3D hardcover mockup of Imbalanced Data: Myths, Mistakes and Modern Solutions
Book cover mockup for Imbalanced Data: Myths, Mistakes and Modern Solutions displayed on a tablet
3D hardcover mockup of Imbalanced Data: Myths, Mistakes and Modern Solutions
Book cover mockup for Imbalanced Data: Myths, Mistakes and Modern Solutions displayed on a tablet
MRMR: Maximum Relevance Minimum Redundancy
Feature importance and feature selection: 2 sides of the same coin
Embedded methods: pros vs cons and workflow
For large datasets with correlated features, MRMR works well — but only for linear relationships
After running feature selection on one-hot encoded variables, finding only 2 out of 10 categories matter is useful information
Feature selection doesn't just improve accuracy — it makes models faster, cheaper, and safer to deploy
Aims of feature selection: improve model performance; speed and cost efficiency; clarify model predictions.
Boruta selects ALL relevant features — not just the optimal model subset