Unsupervised Financial Fraud Detection Using Low-rank Recovery

The goal of this research is to give an effectively unsupervised financial fraud detection method, in which low rank recovery method is used to exploit intrinsic relationship of data samples. The benefits of our methods have several folds: (1) compared with classification-based methods, our methods can deal with the imbalance problem effectively; (2) by using low rank recovery, our methods can exploit intrinsic relationship of data samples effectively; (3) compared with PCA-based methods, the prior used in our methods are more reasonable for the case of financial fraud detection.

Faculty Supervisor:

Xianta Jiang

Student:

Partner:

NASDAQ Canada Inc

Discipline:

Computer science

Sector:

Information and Communications Technology; Health and Related Sciences & Technology; Finance and Insurance

University:

Memorial University of Newfoundland

Program: