PFCT And Seoul National University Research Team Reveal Joint AI Credit Evaluation Study Results

AI technology finance company PFCT announced on the 12th the results of a 10‑month joint study conducted with a research team from Seoul National University to extend AI credit evaluation methodologies and explore various experimental possibilities.

The joint research was carried out with the aim of creating a more sophisticated AI credit evaluation model by comparing and testing various algorithms based on de-identified real financial data. Because the data used for credit evaluation mainly consists of structured information in which each item has its own meaning, tree-based models with strong statistical interpretability have typically been used. In contrast, deep learning-based models are characterized by their ability to learn complex patterns that are difficult to capture statistically, so the two research teams experimentally examined how deep learning-based approaches operate in the field of credit evaluation.

The researchers examined the characteristics and constraints that arise when applying the Transformer structure, one of the deep learning-based approaches, to credit evaluation. Transformers are a learning architecture that has mainly been used in areas such as natural language processing, where the order of data is important. Accordingly, the team focused on analyzing what learning characteristics this structure exhibits in a credit evaluation data environment where there is no concept of word order.

The study found that, under the same data conditions, performance indicators and learning characteristics differed depending on the model architecture. In some experiments, there was a tendency for improvement, compared with conventional methods, in the KS indicator for distinguishing high-risk customers and in performance for capturing lower (low-credit) segments. Through this, the researchers examined the scope and limitations of applying Transformer-based learning methods to credit evaluation. The study is significant in that it secured comparative benchmarks and empirical data for model design with a view to future practical application and for further research.

This joint research project also served as a process for reviewing the advancement direction of AIRPACK, the AI risk management solution that PFCT supplies to financial institutions in Korea and abroad. AIRPACK provides various functions, such as AI credit evaluation and risk strategy derivation and validation, in a modular form. Through this study, PFCT was able to accumulate experimental validation and research data to explore extensions of its existing credit evaluation algorithm architectures.

Moon Byoungro, Professor at the Department of Computer Science and Engineering at Seoul National University, said, "This study is meaningful in that it verified various algorithmic approaches to credit evaluation in a real data environment and confirmed their applicability in the financial field," adding, "By reviewing deep learning-based methodologies in line with the characteristics of credit evaluation data, we have presented benchmarks for future practical applications."

Lee Suhwan, CEO of PFCT, said, "This industry-academia collaboration is meaningful in that it treated AI credit evaluation not as a fixed technology, but as a research domain to be continuously expanded and validated," and added, "Based on these results, which combine algorithm research in academia with real-world data from the industry, we plan to continue research and validation on AI credit evaluation technology."

This content was produced with the assistance of AI translation services.