Deep Learning Framework Evaluates Corporate Debt Maturity Risks Across

A research team in China has published an interpretable deep-learning approach to assess corporate debt maturity risk using multidimensional company records. The study, in Financial Innovation, introduces DFFSR (data fusion-based feature selection and reconstruction), which combines internal feature gating, representation compression, and a predictive neural network. It uses a CancelOut layer attached to a stacked autoencoder to map governance, leverage, liquidity, and profitability indicators into a lower-dimensional embedding before generating forecasts. Testing on A-share listed Chinese firms from 2000 to 2023 shows higher predictive accuracy than baseline econometric and machine-learning methods, while also ranking which managerial and financial factors drive mismatches. The work links findings to China’s heavy use of short-term borrowing for long-term fixed-asset investment.