Handling Unbalanced Datasets with Class weights

This repository explains a way to handle unbalanced dataset in deep learning models. The approach helps to get some performacne gain that has been illustrated with MNIST digit classification dataset.

Requirements

1. Numpy
2. Scikit-learn
3. Keras with tensorflow backend

Contents

1. Details on unbalanced dataset and class weight
2. Calculation of class weights from MNIST datast
3. Experiments and performance comparision with and without class weights

Please notify in the issue section if you find any issue in the experiment and also let me know if you can enhance the performance of any of your deep learning model using this approach.