Mitigating the Effects of Class Noise Using Two-Level Filtering Learner Algorithm
Eman Allogmani, Darsana P. Josyula · 2022
Learning with label noise imposes several consequences on learning algorithms. Since label noise causes learning algorithms to learn from corrupted data, the performance of the machine learning models drop enormously, thereby reducing prediction performance. This paper implements an integrated noise removal and training approach to filter out outliers during classifier training in order to improve the performance of the resulting classification models. The results demonstrate that the method is effective and can enhance classification accuracy significantly.