MODELING AN INTELLIGENT SYSTEM FOR FORECASTING AND INTERPRETING THE FINANCIAL CONDITION OF AN ENTERPRISE USING EXPLAINABLE ARTIFICIAL INTELLIGENCE METHODS

Abstract

The gradual digitalization of the business environment and macroeconomic instability necessitate the implementation of predictive analytics tools for assessing the financial stability of business entities. Relevant machine learning (ML) algorithms, particularly ensembles of decision trees, provide high accuracy in classification and risk prediction but operate as “black boxes”. This significantly limits their practical application in corporate risk management and auditing due to a lack of transparency and the inability to justify the logic behind decision-making.

The paper proposes a conceptual mathematical model of an intelligent system for forecasting and interpreting enterprise financial indicators. The model combines the high predictive capability of additive ensembles with Explainable Artificial Intelligence (XAI) methods. The proposed model consists of three interconnected conceptual blocks. The first block forms the predictive core of the system based on a regularized ensemble of decision trees that minimizes the loss function under the nonlinear characteristics of financial datasets. The second block is based on cooperative game theory and the TreeSHAP methodology, enabling the local decomposition of the contribution of each financial indicator to the final prediction, as well as the calculation of the aggregate global importance of factors over a long-term horizon. The third block implements an online monitoring component based on an original reliability indicator that accounts for both the point prediction error of the model and the structural stability of local Shapley contribution vectors relative to their historical profile.

Experimental validation of the system was performed using synthetic financial statements of enterprises. To ensure full reproducibility of the research, the system source code was deployed in an open GitHub repository, allowing analysts to adapt the framework parameters to the industry-specific characteristics of particular local datasets. The integration of XAI tools and the reliability assessment component enables complex nonlinear models to be transformed into transparent decision-support systems while reducing subjective risks in financial analysis.