Research on Financial Fraud Detection Based on Convolutional Neural Networks 

Ma Chi  Zhang Hongxia

Dalian University, Dalian Liaoning, 116622;

Abstract: With the continuous development of financial markets, financial fraud means are growing more and more complex, more and more hidden. Traditional identification methods that rely solely on a single financial data cannot fully identify potential fraud activities. In order to enhance the accuracy of fraud identification, this paper proposes a multi-modal financial fraud classification model that combines text and structured financial data. Firstly, in the feature extraction stage, the model of this paper uses a combination of BERT ( bidirectional encoder representation of transformers ) and BiLSTM ( bidirectional long-term and short-term memory network ) to deeply extract the text semantic features of management discussion and analysis (MD&A) in the annual reports of listed companies; At the same time, it extracts features from standardized financial data through a fully connected layer. Secondly, in the feature fusion stage, the model adds the feature vectors from the two modalities and introduces the Convolutional Neural Network (CNN) to complete the deep fusion, so as to obtain a more discriminative feature representation. Through the empirical analysis of the samples of Chinese listed companies from 2015 to 2023, it is verified that the multi-modal classification model proposed in this paper can improve the identification and classification performance of financial fraud risk.

Keywords: Financial Fraud Detection; Multimodality; BERT; Convolutional Neural Network

References

[1]Huang Z G, Liu J J, Lin C Y. Comparative study on frontier methods of financial statement fraud identification of listed companies based on machine learning[J]. Journal of Systems Science and Mathematical Sciences, 2020, 40(10): 1882-1900.

[2]Persons O S. Using financial statement data to identify factors associated with fraudulent financial reporting[J]. Journal of applied business research, 1995, 11(3): 38.

[3]Beasley M S. An empirical analysis of the relation between the board of director composition and financial statement fraud[J]. Accounting review, 1996: 443-465.

[4]Loughran T, McDonald B. When is a liability not a liability? Textual analysis, dictionaries, and 10‐Ks[J]. The Journal of finance, 2011, 66(1): 35-65.

[5]Brown N C, Crowley R M, Elliott W B. What are you saying? Using topic to detect financial misreporting[J]. Journal of Accounting Research, 2020, 58(1): 237-291.

[6]Wang R, Asghari V, Hsu S C, et al. Detecting corporate misconduct through random forest in China’s construction industry[J]. Journal of cleaner production, 2020, 268: 122266.

[7]Liu R, Huang J, Zhang Z. Tracking disclosure change trajectories for financial fraud detection[J]. Production and Operations Management, 2023, 32(2): 584-602.