Supply Chain Finance Risk Prediction Using Temporal Fusion Transformer on Transaction Time-Series

Supply Chain Finance (SCF) is a process of controlling the flow of capital between the stakeholders in a supply chain, and correct prediction of risk is necessary to maintain financial stability and safety of investments. Utilizing transaction time-series information is also a useful way to get the time-related patterns that can be used to predict financial risk. Nevertheless, the current techniques like classical statistical models and simple machine learning techniques are not able to explain long-range correlations and dynamic time, resulting in low risks prediction accuracy. In order to overcome these drawbacks, we introduce a new model, namely SCF-TFT (Supply Chain Finance - Temporal Fusion Transformer), which combines a Temporal Fusion Transformer model with more effective feature selection and time-related attention. The methods used in the proposed SCF-TFT framework include variable selection networks, gating mechanisms, and explainable multi-horizon forecasting, which will adequately model complex temporal relationships and enhance robustness of predictions. In the supply chain field, this technique is used on actual data on transaction time-series to detect possible default risks and examine the financial condition of suppliers over a period of time. The experimental findings indicate that SCF-TFT is superior to baseline models and can be more precise, superior in recall, and F1-scores in risk prediction. Also, the model gives interpretable outputs that identify critical time drivers to affect financial risk, and therefore the model is useful in decision-making of those in finance and supply chain management.

Hali tarjima qilinmagan