cifrus 1.0.4

Official implementation of Class-invariante Feature Range Uniform Sampling (CiFRUS)

Class-invariant Feature Range Uniform Sampling (CiFRUS)

Introduction

This is the official implementation of the tabular data augmentation method described in the following KDD 2024 paper: A Novel Feature Space Augmentation Method to Improve Classification Performance and Evaluation Reliability. The core augmentation method is class-invariant and supports majority-based prediction of unlabeled instances. Each unlabeled instance can be expanded into a set of augmented instances, followed by classifier prediction and aggregation of the predicted labels or class probabilities.

Installation

Installing with Pip

CiFRUS can be installed from PyPI with pip:

python -m pip install cifrus

The PyPI package only contains the core augmentation module. To acquire the datasets and code for conducting experiments, please fork this repository.

From source

cifrus.py can be directly copied to the project repository.

Usage

CiFRUS is compatible with scikit-learn and imbalanced-learn. Given feature matrix X where each row is a sample and corresponding class labels y, augmentation can be performed as follows:

from cifrus.cifrus import CiFRUS
cfrs = CiFRUS()
X_resampled, y_resampled = cfrs.fit_resample(X, y)

By default, CiFRUS augmentation increases the number of samples in the majority class five-fold, and balances the minority classes.

The majority-based prediction of CiFRUS can be used by passing the classifier's predict_proba() function as a parameter to the resample_predict_proba function of the CiFRUS object, as demonstrated below:

# any sklearn compatible classifier
from sklearn.ensemble import RandomForestClassifier
clf = RandomForestClassifier() 
clf.fit(X_resampled, y_resampled)

# X_test is resampled, and the probabilities are aggregated internally
y_pred = cfrs.resample_predict_proba(clf.predict_proba, X_test)

License

This work is licensed under the Creative Commons Attribution 4.0 International License.

Citation

If you use CiFRUS in your scientific research, please consider citing us:

@inproceedings{saimon2024novel,
    author = {Saimon, Sakhawat Hossain and Najnin, Tanzira and Ruan, Jianhua},
    title = {A Novel Feature Space Augmentation Method to Improve Classification Performance and Evaluation Reliability},
    doi = {10.1145/3637528.3671736},
    booktitle = {Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining},
    pages = {2512–2523},
    year = {2024},
}