KPMG AZSA & Co., together with Hitotsubashi University, developed a technology to detect fraudulent accounting at the account level, and obtained a patent (No. 6678665) on February 28, 2020.
KPMG AZSA has already used a fraud risk detection model that utilizes AI and machine learning in its audit operations. By using the patent technology here, it will be possible to calculate a more precise degree of abnormalities at the account level in a probabilistic manner in the future, which will lead to early detection and response to fraud risk.
This patent technology uses a method called Graphical Lasso to input the accounting information of the verification target company into a model built by learning the relationship between accounts from data, and calculates the degree of abnormality of each account are doing. This model adopts a hybrid structure that includes two methods: (1) Supervised Learning, which learns by giving information on the presence or absence of past fraud, and (2) Unsupervised Learning, which learns without giving the same information. It is possible to detect fraudulent accounting with a small number of cases in the past.