A data-centric approach to building ai models for determining the credit rating of fintech company clients based on open banking

This paper implements a data-centric approach to training and testing models that assess a client’s credit rating based on their transactions that are obtained using open banking. Well-defined logic rules were used to create synthetic data, and additional synthetic datasets were generated using GANs. The training and testing of a machine learning model (Random Forest) and neural network models implemented in TensorFlow and PyTorch with identical architectures were performed. The process of adapting the models to data updates is examined. Data clustering was used to identify patterns of behaviour among clients who did not repay their loans. The quality of the generated data was evaluated. Classification results are presented depending on the type of model, the amount of data, and the consideration of client behaviour patterns. The results showed that Random Forest models provide the highest accuracy regardless of the type of dataset. The accuracy of training neural networks depends significantly on the structure of the datasets