TY - GEN
T1 - MUVIR
T2 - 24th International Joint Conference on Artificial Intelligence, IJCAI 2015
AU - Zhou, Dawei
AU - He, Jingrui
AU - Candan, K. Seluk
AU - Davulcu, Hasan
PY - 2015
Y1 - 2015
N2 - Rare category detection refers to the problem of identifying the initial examples from under-represented minority classes in an imbalanced data set. This problem becomes more challenging in many real applications where the data comes from multiple views, and some views may be irrelevant for distinguishing between majority and minority classes, such as synthetic ID detection and insider threat detection. Existing techniques for rare category detection are not best suited for such applications, as they mainly focus on data with a single view. To address the problem of multi-view rare category detection, in this paper, we propose a novel framework named MUVIR. It builds upon existing techniques for rare category detection with each single view, and exploits the relationship among multiple views to estimate the overall probability of each example belonging to the minority class. In particular, we study multiple special cases of the framework with respect to their working conditions, and analyze the performance of MUVIR in the presence of irrelevant views. For problems where the exact priors of the minority classes are unknown, we generalize the MUVIR algorithm to work with only an upper bound on the priors. Experimental results on both synthetic and real data sets demonstrate the effectiveness of the proposed framework, especially in the presence of irrelevant views.
AB - Rare category detection refers to the problem of identifying the initial examples from under-represented minority classes in an imbalanced data set. This problem becomes more challenging in many real applications where the data comes from multiple views, and some views may be irrelevant for distinguishing between majority and minority classes, such as synthetic ID detection and insider threat detection. Existing techniques for rare category detection are not best suited for such applications, as they mainly focus on data with a single view. To address the problem of multi-view rare category detection, in this paper, we propose a novel framework named MUVIR. It builds upon existing techniques for rare category detection with each single view, and exploits the relationship among multiple views to estimate the overall probability of each example belonging to the minority class. In particular, we study multiple special cases of the framework with respect to their working conditions, and analyze the performance of MUVIR in the presence of irrelevant views. For problems where the exact priors of the minority classes are unknown, we generalize the MUVIR algorithm to work with only an upper bound on the priors. Experimental results on both synthetic and real data sets demonstrate the effectiveness of the proposed framework, especially in the presence of irrelevant views.
UR - http://www.scopus.com/inward/record.url?scp=84949808496&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=84949808496&partnerID=8YFLogxK
M3 - Conference contribution
AN - SCOPUS:84949808496
T3 - IJCAI International Joint Conference on Artificial Intelligence
SP - 4098
EP - 4104
BT - IJCAI 2015 - Proceedings of the 24th International Joint Conference on Artificial Intelligence
A2 - Wooldridge, Michael
A2 - Yang, Qiang
PB - International Joint Conferences on Artificial Intelligence
Y2 - 25 July 2015 through 31 July 2015
ER -