Novelty detection algorithm for data streams multi-class problems

Elaine R. Faria, João Gama, André C. P. L. F. Carvalho
2013 Proceedings of the 28th Annual ACM Symposium on Applied Computing - SAC '13  
Novelty detection has been presented in the literature as one-class problem. In this case, new examples are classified as either belonging to the target class or not. The examples not explained by the model are detected as belonging to a class named novelty. However, novelty detection is much more general, especially in data streams scenarios, where the number of classes might be unknown before learning and new classes can appear any time. In this case, the novelty concept is composed by
more » ... nt classes. This work presents a new algorithm to address novelty detection in data streams multi-class problems, the MINAS algorithm. Moreover, we also present a new experimental methodology to evaluate novelty detection methods in multi-class problems. The data used in the experiments include artificial and real data sets. Experimental results show that MINAS is able to discover novelties in multi-class problems.
doi:10.1145/2480362.2480515 dblp:conf/sac/FariaGC13 fatcat:au23wrd7mbatxaujtfwn2swb44