Fuzzy ARTMAP Network and Clustering for Streaming Classification under Emerging New Classes

Streaming classification with emerging new classes is a challenging problem receiving extensive attention. To date, most approaches simply detect novel classes without further recognizing them, and ignore conceptual drift and conceptual evolution problem. Therefore, in this paper, we propose an effective semi-supervised framework that combines ARTMAP neural network and clustering method to detect and identify simultaneously multiple known and unknown classes in data streams and update the network online. Experiments have been conducted on four benchmark datasets with different data forms including MNIST, CIFAR10, network attack analysis, and geospatial information of forests. The empirical evaluation shows effectiveness of the proposed approach, whose results are much better than many previous studies.