Using an Ensemble LSTM Model for Financial Statement Fraud Detection

Authors

Yahui Sun, Fudan UniversityFollow
Yue Wu, Fudan UniversityFollow
Yunjie (Calvin) Xu, Fudan UniversityFollow

Abstract

Financial statement fraud is a serious problem in global financial markets. We propose an ensemble long short-term memory (LSTM) model to detect financial statement fraud. Using a data set with 2,621 fraudulent reports and 11,144 non-fraudulent reports, we extract three categories of features: financial features from annual reports, linguistic features from the management discussion and analysis (MD&A) section of annual reports, and paragraph vectors from the MD&A section. The ensemble LSTM model feeds each of the three categories of features into a classic LSTM, and integrates the output layers of the three LSTM models with a multilayer perceptron (MLP). We benchmark the performance of our model against a random forest model. Our model outperforms the baseline model and identifies more effective input features.

Recommended Citation

Sun, Yahui; Wu, Yue; and Xu, Yunjie (Calvin), "Using an Ensemble LSTM Model for Financial Statement Fraud Detection" (2020). PACIS 2020 Proceedings. 144.
https://aisel.aisnet.org/pacis2020/144