AI-Augmented Early Warning Models for Commercial and SME Segments: Leveraging Unstructured Data and Time-Series Analytics

The Early Warning System is a critical application that uses advanced AI algorithms to monitor large amounts of data to detect potential payment difficulties f or c lients. In this paper, we present a novel approach to generate time-series indicators from new data sources such as Trade Registry Gazette announcements and records of the number of days of delinquency. Our method leverages machine learning algorithms such as Extreme Gradient Boosting (XGBoost) to identify hidden patterns and trends in these data sources, thus overcoming the problems with the traditional rule-based models. By analyzing this data, we can identify time-series anomalies and predict the likelihood of 10 days of delinquency 6 months ahead for the clients in the SME (Small and medium-sized enterprises) and Commercial segments. Our analysis demonstrates that our approach can achieve high accuracy in delinquency prediction, enabling financial i nstitutions t o t ake p roactive m easures to prevent potential losses. Our proposed method can enrich credit monitoring systems and enhance the ability of financial institutions to mitigate financial risk.