This repository contains a Python implementation of the Variance Reduction Synthetic Minority Over-sampling Technique (VR-SMOTE), which is an enhanced version of the Synthetic Minority Over-sampling Technique (SMOTE). The VR-SMOTE aims to improve dataset balancing by strategically oversampling minority class instances in a way that reduces variance and prevents overfitting.
Traditional SMOTE techniques can sometimes lead to overfitting when they oversample the minority class without considering the variance within the data. VR-SMOTE addresses this by first clustering the minority class samples using K-Nearest Neighbors (KNN) and then calculating a relative variance score (RV score) for each cluster. Clusters with high and medium variance are then oversampled, while clusters with low variance are not, in an effort to maintain a balanced variance across the dataset.
Before running this code, you'll need to install the required Python libraries. You can do this by running:
pip install -r requirements.txt
This will install the following packages:
- numpy
- scikit-learn
- imbalanced-learn
The main function in vr_smote.py can be used as follows:
from vr_smote import vr_smote
# Load your data
data = # ... (numpy array with shape [n_samples, n_features + 1])
# The last column in `data` should be the class/target variable
# Apply VR-SMOTE
balanced_data = vr_smote(data)Please ensure your data is preprocessed accordingly, with numerical features and a binary target variable indicating class membership.
To run an example with synthetic data, you can use the script example.py:
python example.py
This will generate a synthetic dataset, apply VR-SMOTE, and print the shapes of the original and balanced datasets.
Contributions to this project are welcome. Please fork the repository and submit a pull request with your changes.
This project is open-sourced under the MIT License. See the LICENSE file for details.
If you use this method in your research, please cite the following paper: Karkaria, V., Chen, J., Siuta, C., Lim, D., Radelescu, R., & Chen, W. (2023). A MACHINE LEARNING BASED TIRE LIFE PREDICTION FRAMEWORK FOR INCREASING LIFE OF COMMERCIAL VEHICLE TIRES. Journal of Mechanical Design, 1-21.