New Book Reveals Imbalanced Data Truths: Data and Models Matter More Than Balancing

Al_Grigor · x · 2026-08-12

Soledad Galli released a new book, Imbalanced Data: Myths, Mistakes and Modern Solutions. The author tested widely accepted approaches for imbalanced data across 37 datasets, including SMOTE, over/undersampling, and evaluation metrics.

The core conclusion is that data size, class overlap, model choice, and evaluation methods matter more than class imbalance itself. Most techniques recommended for imbalanced data work not because they balance classes, but because they fix other underlying issues.

Original post →

More from Research

Research channel →