Gradient Boosting for Defect Detection is the application of gradient boosted tree models (XGBoost, LightGBM, CatBoost) to identify and classify wafer defects — sequentially building trees that focus on the hardest-to-classify examples for superior detection accuracy.

How Does Gradient Boosting Work?

• Sequential: Each new tree corrects the errors of the previous ensemble.
• Gradient: Fits trees to the negative gradient of the loss function (residuals).
• Regularization: Learning rate, max depth, and L1/L2 penalties prevent overfitting.
• XGBoost: The dominant implementation, with efficient handling of sparse data and missing values.

Why It Matters

• Best Tabular Performance: Gradient boosting consistently wins Kaggle competitions and industrial benchmarks on tabular data.
• Defect Classification: Classifies defect types from SEM images, wafer maps, or process data.
• Class Imbalance: Handles the severe class imbalance common in defect data (rare defects vs. many good samples).

Gradient Boosting is the premier ML algorithm for structured fab data — sequentially correcting errors for the best defect detection accuracy on tabular process data.