Data Balancing

Endpoint: POST /data-balancing

Request Parameters

File Upload

• file (required): The file to be processed. Supported formats include CSV, Excel, etc.

Form Parameters

• target_column (required): The target column to balance the dataset on.
• method (optional, default: oversampling): The balancing method to apply. Supported values:
  • oversampling: Increases the number of samples in the minority class.
  • undersampling: Decreases the number of samples in the majority class.
  • hybrid: Combines oversampling and undersampling techniques.
  • class_weighting: Adjusts class weights to handle imbalance.
• oversampling_method (optional, default: SMOTE): The oversampling method to use if method is oversampling. Supported values:
  • SMOTE: Synthetic Minority Over-sampling Technique.
  • ADASYN: Adaptive Synthetic Sampling.
  • RandomOverSampler: Random oversampling.
• k_neighbors (optional, default: 5): The number of neighbors to use for oversampling methods like SMOTE and ADASYN.
• random_state (optional, default: 42): Random seed for reproducibility.
• undersampling_method (optional, default: RandomUnderSampler): The undersampling method to use if method is undersampling. Supported values:
  • RandomUnderSampler: Random undersampling.
  • ClusterCentroids: Undersampling using cluster centroids.
  • TomekLinks: Removes Tomek links to clean the dataset.
• sampling_strategy (optional, default: 0.8): The sampling strategy for balancing. Can be a float or a dictionary specifying the desired ratio of classes.
• class_weights (optional, default: balanced): The class weighting strategy if method is class_weighting. Supported values:
  • balanced: Automatically adjusts weights inversely proportional to class frequencies.
  • custom_weights: Allows custom weights (not implemented in this version).

Processing Logic

1. Read the uploaded file: The file is converted into a Pandas DataFrame.
2. Validate the target column: Checks if the specified target column exists in the DataFrame.
3. Encode categorical columns: Converts categorical columns into numerical values using label encoding.
4. Apply the selected balancing method:
   • Oversampling:
     • SMOTE: Generates synthetic samples for the minority class.
     • ADASYN: Generates synthetic samples adaptively.
     • RandomOverSampler: Randomly duplicates samples from the minority class.
   • Undersampling:
     • RandomUnderSampler: Randomly removes samples from the majority class.
     • ClusterCentroids: Reduces the majority class by generating centroids.
     • TomekLinks: Removes Tomek links to clean the dataset.
   • Hybrid: Combines oversampling and undersampling using SMOTEENN.
   • Class Weighting: Adjusts class weights to handle imbalance without modifying the dataset.
5. Decode categorical columns: Converts numerical values back to their original categorical values.
6. Return the balanced dataset: The balanced dataset is provided for download with the filename appended with _{method}_balanced.

Example Request

curl -X POST "http://localhost:8000/data-balancing" \
-F "file=@data.csv" \
-F "target_column=target" \
-F "method=oversampling" \
-F "oversampling_method=SMOTE" \
-F "k_neighbors=5" \
-F "random_state=42" \
-F "sampling_strategy=0.8"