An Explainable AI-Based Framework for Predicting Construction Firm Profitability and Capital Structure Considering Supply-Chain Volatility

초록

Traditional financial models often fail construction firms by ignoring industry-specific volatility. This study proposes an Explainable AI (XAI) framework to predict Return on Assets (ROA) and Debt-to-Equity (D/E) ratios using KOSPI and KOSDAQ data (2015-2024). It compares a baseline financial dataset (Dataset 1) with an industry-augmented version (Dataset 2) that incorporates the Producer Price Index (PPI) and labor wages. Using ANN, LSTM, and RF, the study found that the RF model achieved the highest predictive power (R2=0.7978). Notably, Dataset 2 improved accuracy significantly, reducing the Mean Absolute Error (MAE) for ROA by approximately 56%. SHAP analysis revealed that rising rebar and steel prices (PPI_RB) negatively impact both profitability and stability. In contrast, the concrete price index (PPI_CO) showed a positive influence, reflecting firms' ability to manage costs or adjust prices. Crucially, material costs proved far more influential than labor costs. These findings indicate that supply-chain and raw-material-price indicators are salient predictors, suggesting that management may prioritize monitoring and mitigating supply-chain volatility, while causal validation remains beyond the scope of this study. Ultimately, by leveraging an XAI-based interpretation, this study provides a decision-making basis for formulating procurement and hedging strategies using raw material price indices, while also proposing an analytical framework that explains financial performance in the construction industry from a volatility-oriented perspective.

키워드