Liquidity crisis is one of the most critical threats to corporate sustainability and financial stability. The inability of firms to meet short-term obligations can lead to operational disruptions, financial distress, and ultimately bankruptcy. Traditional liquidity prediction models are primarily based on financial ratios and classical statistical techniques. Although these approaches have been widely used in the literature, they often fail to capture complex nonlinear relationships and temporal dependencies embedded in financial statement data. Recent advances in artificial intelligence and deep learning have created new opportunities for developing more accurate predictive models for financial risk assessment.This study proposes a hybrid deep learning framework based on Transformer architecture and XGBoost algorithm for predicting corporate liquidity crises. The research utilizes financial statement data from firms listed on the Tehran Stock Exchange over the period 2013–2023. Financial indicators including liquidity ratios, profitability measures, leverage ratios, activity ratios, and cash flow variables were employed as model inputs. The Transformer network was used to extract temporal and structural patterns from sequential financial data, while XGBoost served as the final classification layer to identify firms facing liquidity distress.The performance of the proposed model was compared with several benchmark models, including Logistic Regression, Random Forest, Artificial Neural Networks, and Long Short-Term Memory (LSTM) networks. Empirical results indicate that the Transformer-XGBoost hybrid model significantly outperforms alternative approaches, achieving an accuracy of 96.8% and an ROC-AUC score of 0.983. Furthermore, operating cash flow, current ratio, debt ratio, and cash conversion cycle were identified as the most influential predictors of liquidity distress.The findings demonstrate that integrating deep representation learning with advanced ensemble classification techniques can substantially improve the early detection of liquidity crises. The proposed framework provides practical implications for financial managers, investors, creditors, and regulatory authorities by enhancing early warning systems and supporting more informed financial decision-making.