EG-SMOTE algorithm is an enhancement of G-SMOTE algorithm for efficient resampling that addresses the limitations of
- Synthesizing noisy minority samples
- Overfitting due to extreme synthesis of minority samples
- Improper synthesis along the borderlines.
Two types of imbalances are handled in EG-SMOTE;
- between-class imbalance
- imbalance is handled with improved resampling methodology
- within-class imbalance
- rectified by applying the k-means clustering algorithm and then applying resampling within clusters.
Requirement Modules
- Python 3
- numpy
- pandas
- imblearn
- Scikit-learn
Getting Started
- A sample dataset can be found in the Experiment/data/NSLKDD-mini.csv
- The model architecture can be found in egsmote.py.
- And you can try the demo experiment by running Experiment/demo.py file.