SMOTENCOptions<T>
Options & Configuration · AiDotNet.Models.Options
Configuration options for SMOTE-NC (Synthetic Minority Over-sampling Technique for Nominal and Continuous features), a k-NN based oversampling method that generates synthetic minority samples by interpolating between existing ones.
Inheritance: ModelOptions → NeuralNetworkOptions → ModelHyperparameterOptions → FinancialNeuralNetworkOptions → RiskModelOptions<T> → SMOTENCOptions<T>
For Beginners
SMOTE-NC helps when your data is imbalanced (e.g., 99% normal transactions, 1% fraud).
How it works:
- Find each minority sample’s k nearest neighbors (similar samples)
- Pick one neighbor randomly
- Create a new sample “in between” the original and the neighbor
- For numbers: pick a random point on the line between them
- For categories: use the most common category among the neighbors
This produces new minority samples that are realistic because they’re based on actual data relationships, not just random copies.
Example:
How It Works
SMOTE-NC extends the original SMOTE algorithm to handle mixed-type data:
- Continuous features: Linear interpolation between a sample and its k-nearest neighbor
- Categorical features: Majority vote among the k-nearest neighbors
- Distance metric: Euclidean for continuous + Value Difference Metric for categoricals
Reference: “SMOTE: Synthetic Minority Over-sampling Technique” (Chawla et al., 2002)
Constructors
| Constructor | Summary |
|---|---|
SMOTENCOptions() |
Properties
| Property | Summary |
|---|---|
int K { get; set; } |
Gets or sets the number of nearest neighbors to consider. |
int LabelColumnIndex { get; set; } |
Gets or sets the index of the label column for identifying minority class. |
int MinorityClassValue { get; set; } |
Gets or sets the minority class value to oversample. |