1.5 Class imbalance and synthetic data

Key points

1. Class imbalance means one label is much rarer than another. A balanced class weight makes errors on the rare class cost more during training.

   What NVIDIA says (2)

   “The “balanced” mode uses the values of y to automatically adjust weights inversely proportional to class frequencies in the input data as n_samples / (n_classes * np.bincount(y)) .”

   — cuML API: LogisticRegression

   “If 'balanced' , class weights are computed from the training labels.”

   — cuML API: RandomForestClassifier

2. A stratified split keeps each class's share the same in every part. Without it, a small test set may contain few or no rare cases.

   What NVIDIA says (1)

   “If not None, data is split in a stratified fashion, using this as the class labels.”

   — cuML API: train_test_split

3. Synthetic data is artificial data made by rules, simulations or generative models. It can fill gaps in rare classes and protect privacy.

   What NVIDIA says (2)

   “Data Quality : Real-world datasets can be imbalanced, which can result in biased outputs from generative models and ML models.”

   — NVIDIA Glossary: Synthetic Data Generation

   “Data Privacy : Synthetic data helps overcome privacy issues by generating training data that mimic real-world statistics without directly corresponding to individual records.”

   — NVIDIA Glossary: Synthetic Data Generation

4. Accuracy is the share of correct predictions. On imbalanced data it hides poor results on the rare class; a confusion matrix shows them.

   What NVIDIA says (2)

   “Accuracy classification score.”

   — cuML API: accuracy_score

   “Compute confusion matrix to evaluate the accuracy of a classification.”

   — cuML API: confusion_matrix

Key terms

• Class imbalance: When one class is much rarer than the others in the data.
• Stratified split: A train-test split that keeps the same class ratio in both parts.
• Synthetic data: Generated data that mimics real data, used to fill gaps or protect privacy.
• Accuracy: The share of predictions that are correct.

Try it

• Confusion matrix lab: precision, recall and F1

Sample question

Only 2% of transactions are fraud. Which cuML classifier setting gives the rare class more weight automatically?