With the increasingly fierce market competition, enterprises have increasingly adopted credit sales strategies to expand the market, enhance inventory turnover rates and reduce storage costs. However, the popularity of credit sales also brings about a large proportion of accounts receivable and bad debt risk, which seriously threatens the capital liquidity of enterprises. Motivated by the powerful capabilities of deep neural networks (DNN), we propose a DNN-SHAP model to predict the repayment time of accounts receivable, including after overdue default based on historical repayment records. Visualization and interpretability are also implemented based on shapley value to explore the primary factors and underlying mechanisms of overdue repayment behavior. Furthermore, we present a new dataset derived from actual B2B order invoice scenarios and evaluate the performance of proposed model on this dataset. By comparing traditional machine learning models such as support vector machines and random forests with our model, the results show that the DNN-SHAP model has significant advantages in accuracy. This finding suggests that the DNN model can effectively capture key features in repayment timing prediction, offering a new perspective and tool for credit risk management and the forecasting of overdue repayments.