Bagging Folds using Synthetic Majority Oversampling is a novel approach designed to address the challenges of binary imbalanced data classification. It is a meta framework that incorporates techniques such as data partitioning, threshold optimization, oversampling of the majority class and classifier ensemble.
In the initial stage, BaggFold divides the data set into perfectly balanced folds, each containing an equal number of minority and majority instances. The majority of instances are selected without replacement.
In the second step, each fold is uniquely assigned to a base classifier. Subsequently, each classifier fine-tunes its decision threshold using Youden's J statistics.
The final step involves training in a modified bagging fashion, where BaggFold performs both training and inference concurrently. This approach enhances performance and reduces inference times, contributing to the overall efficiency of the framework.
This repository also implements two SMOTE techniques:
- Center Point SMOTE (CP-SMOTE),
- Inner and Outer SMOTE (IO-SMOTE).
Both are presented in a paper Two novel smote methods for solving imbalanced classification problems by Yuan Bao & Sibo Yang.
- Create virtual environment
conda create --name baggfold python=3.8 - Activate created virtual environment
conda activate baggfold- Within virtual env run
python install.py- Run script (3 BaggFold variants on multiple datasets)
python main.pyIn this directory there is a BaggFold implementation.