The disparity between the official exchange rate and the market rate has created opportunities for currency manipulators to exploit the system. On the other hand, importers of goods are evaluated by the national banking network regardless of their past foreign exchange and financial performance, and collateral is required from them in the form of domestic currency. The main objective of this study is to design an intelligent deep learning model for risk assessment and collateral determination for importers, in such a way that the final model, with the highest level of accuracy and precision, is capable of analyzing large-scale performance data. For this purpose, operational data were first clustered using the K-means method, and the results of the clustering were then used as input for classification models including Random Forest, XGBoost, and Keras Sequential neural networks. The performance of each model was evaluated using the F-measure index. Finally, the combined K-means–RNN model was selected as the best-performing model with the highest accuracy and precision. Using this model, customers were classified into three risk categories, and an optimal mix of cash and non-cash collateral was determined for each group.
The findings indicate that the proposed model is capable of effectively classifying customers based on their performance history and can serve as a robust tool for credit and foreign exchange risk management in the banking sector.