Analysis of financial fraud based on manager knowledge graph
Shigang Wen, Jianping Li, Xiaoqian Zhu, Mingxi Liu
Abstract
Financial fraud can result in devastating consequences for the stability of a firm, as well as considerable losses to shareholders, industry, and even the whole market. The existing studies for fraud detection mainly rely on traditional data sources, which utilize limited information from financial statements. In this paper, we design a novel knowledge graph (KG) framework that aims to enhance financial fraud detection by discovering knowledge from the relationship between the managers and the associated institutions. Using four machine learning algorithms, we evaluate the classification performance of the proposed framework. The empirical analyses show that the support vector machine (SVM) model achieves the best testing performance among all classifiers by applying a training set of financial ratios and firms’ topological features derived from the KG. In particular, the precision is significantly improved by taking the topological features into account. These findings suggest that KG can be effectively used to detect financial fraud.