BalancedSplitter<T>
Models & Types ยท AiDotNet.Preprocessing.DataPreparation.Splitting.Stratified
Balanced splitter that ensures equal representation of each class in train and test sets.
Inheritance: DataSplitterBase<T> โ BalancedSplitter<T>
For Beginners
When you have imbalanced classes (e.g., 90% normal, 10% fraud), this splitter ensures both train and test sets have equal numbers of each class by undersampling the majority classes.
How It Works
How It Works:
- Find the smallest class size
- Sample equally from each class to match the smallest
- Split the balanced data into train/test
When to Use:
- When you want to evaluate model performance without class imbalance effects
- When minority class performance is critical
- For initial model development before handling imbalance in other ways
Caution: This discards data from majority classes. For production, consider SMOTE or other oversampling techniques instead.
Constructors
| Constructor | Summary |
|---|---|
BalancedSplitter(double testSize, bool shuffle, int randomSeed) |
Creates a new balanced splitter. |
Properties
| Property | Summary |
|---|---|
bool RequiresLabels { get; } |
|
string Description { get; } |
Methods
| Method | Summary |
|---|---|
DataSplitResult<T> Split(Matrix<T> X, Vector<T> y) |