An Attributed Network Features Learning Method for Over-Indebtedness Prediction

Over-indebtedness is a financial anomaly, serving as an early warning signal of financial distress in companies. Machine learning provides data-driven cognitive enhancement tools, enabling researchers to accurately predict over-indebtedness. Although existing prediction methods help mitigate the damage caused by over-indebtedness, they often focus solely on internal financial metrics, overlooking the interactions of companies within complex social networks. This isolated perspective may lead to cognitive biases and reduced predictive accuracy. Therefore, this paper proposes an over-indebtedness prediction model based on attributed network feature learning for early warning. Building on previous research, this model incorporates external information, such as interlocking directorate networks and product competition networks, as data sources for feature construction. To build accurate and effective social network structural features, we mimic the human process of integrating and understanding multi-source information in complex environments and propose an attributed network feature learning approach. This approach represents network data through descriptive methods and deep attributed network embedding methods, generating comprehensive features that combine individual and external information. To maximize the model's performance, the study employs a generative classifier, specifically a Bayesian learning approach based on the locally weighted expectation-maximization method, to identify unobservable factors. The study evaluates the model using seven metrics, and the results indicate that the model performs well. This research provides a novel cognitive framework for financial prediction modeling, demonstrating the cognitive value of external information in corporate decision-making.