For any ML project, the importance of high-quality data is paramount. Imbalance in data often causes bias and reduces the robustness of models trained on such data.
At Hamad bin Khalifa University, Dr. Samir B. Belhaouari, Dr. Halima Bensmail, Dr. Atiq Rehman, and I put our heads together to create KNNOR, a novel data augmentation technique that generates artificial data points safely and smartly in order to balance disproportionate data.
Turns out that we were on to something as the algorithm fared better than the state-of-art and enabled classifiers to attain higher accuracy.
K-Nearest Neighbor OveRsampling technique or KNNOR in short is now published in Applied Soft Computing and available to be used as a python library.
Check the library in action at