An Explainable Multi-Level Approach for Financial Fraud Detection

As financial information is the fundamental source of decision-making for numerous stakeholders, investors and regulators, there is a risk that some companies may manipulate it to mislead people and engage in fraudulent activities. However, the growth of capital markets and accelerated advances in information technology have contributed to the increasing sophistication and insidiousness of financial fraud. Consequently, it is becoming increasingly difficult for some well-established models to identify new fraud patterns, making the identification and prevention of financial fraud an even more challenging phenomenon. In this paper, we propose a novel multi-layer architecture model, through the integration of financial features, business features, and strategic features. The framework employs three neural networks for effective feature extraction, and concatenates them together for classification. Additionally, we develop this framework using Explainable Artificial Intelligence (XAI) methods. Our results show that the multi-layer architecture model exhibits better predictive performance than traditional models, and can better meets the requirements of different external stakeholders by providing explanations.