ADASYN (ADAptive SYNthetic sampling) is an improvement over SMOTE that adaptively generates more synthetic samples in regions where minority examples are harder to learn — focusing synthetic data generation on the minority samples near the decision boundary or surrounded by majority samples.

How ADASYN Works

• Density Estimation: For each minority sample, compute the ratio of majority neighbors within $k$ nearest neighbors.
• Difficulty: Samples with more majority neighbors are "harder" — generate MORE synthetic samples near them.
• Adaptive: The number of synthetic samples per minority example is proportional to its local difficulty.
• Smoothing: Normalize the difficulty ratios to obtain sampling weights.

Why It Matters

• Targeted: Unlike SMOTE (which treats all minority samples equally), ADASYN focuses on the hardest regions.
• Decision Boundary: More synthetic samples near the decision boundary = better learned boundary.
• Adaptive: Automatically identifies which minority regions need the most augmentation.

ADASYN is smart SMOTE — adaptively generating more synthetic samples where the minority class is hardest to learn.