This repository provides a Pytorch implementation of the paper "Valid Oversampling Schemes to Handle Imbalance", accepted at Pattern Recognition Letters. We compare accuracy, sensitivity, and specificity of various oversampling methods using the MNIST dataset. [paper link].

To execute the valid oversampling method, please go to the directory lib first and run the main.py file as follows.

python3 main.py --experiment 4 # Valid-oversampling method

The followings are arguments for other experitments.

Experiment 0: No oversampling
Experiment 1: Oversample minor class to be balanced
Experiment 2: Use a cost-sensitive loss function
Experiment 3: SMOTE
Experiment 4: Valid oversampling

[1] Chawla, N.V., Bowyer., K.W., Hall, L.O., Kegelmeyer, W.P., 2002. Smote:Synthetic minority over-sampling technique. Journal of Artificial Intelligence Research 16, 321–357.