To search, Click below search items.


All Published Papers Search Service


Evaluation of Class Noise Impact on Performance of Machine Learning Algorithms


Zahra Nazari, Masooma Nazari, Mir Sayed Shah Danish, and Dongshik Kang


Vol. 18  No. 8  pp. 148-152


Real-world datasets are not perfect and always suffer from noise that may affect classifiers built under the effect of such type of disturbance. Different types of noise are existing in almost any real-world problem, but not always known. Existence of noise decreases the accuracy of a classifier and increases its training time and complexity of the induced model. Most of existing machine learning algorithms have integrated different approaches to enhance their learning abilities in presence of noise, but noise still can make negative impacts. Therefore noise robustness of a classifier is an important issue in noisy environments and should be studied. This paper evaluates the robustness of different machine learning algorithms against class noise. The Equalized Loss of Accuracy (ELA) is the robustness metric which is used in this study. Ten benchmark datasets with 0-20% of noise level are used in experiments and finally ELA results of algorithms are compared.


Noise impact, Classification, Robustness metric, Class noise