Detection and elimination of class noise in large datasets using partitioning filter technique
Btissam Zerhari, Ayoub Ait Lahcen, Salma Mouline · 2016
Class noise elimination in large databases is a real issue in data mining processing. In fact, class noise may sometimes lead to distortion or inaccuracy. So to overcome this problem, many techniques have been proposed. However, most of them don't have the capacity to deal with huge volume. In this context, this paper presents an architecture for class noise detection and elimination in large datasets. This architecture relies on four important levels: i) dividing data into subsets; ii) extracting best rules to predict classes; iii) applying different classifiers to the subsets in order to detect class noise; and finally iv) combine the classifiers results. The validity of the architecture is studied in experimentation, comparing other works with our proposal.