A Novel Structure-based Feature Extraction Approach for Financial Fraud Detection

Abstract

In finance and economic area, financial fraud detection plays an important role for both corporate management and capital market system. Feature extraction is one of the most important procedure in fraudulent firm detection. Current feature extraction approaches pay large amounts of attention on their financial attributes, which have explicitly limited the representation of ‘normal’ pictures of firms, and furthermore, reduced the financial fraud detection performance. Hence it is necessary to search for a better set of functions as features to represent firms more accurately. In this work, notice that the imitation behaviors among firms often happen in business management, while this structure patterns have not been utilized in f inancial fraud detection so far, we extract features under the constraint of both financial characters and structure patterns of firms. We also design three measurements to quantify the structure patterns. Experimental results have shown a great performance of the proposed approach.