This project aims to mitigate heterogeneous DeFi lending risk through machine-learning-based assessment methods and warning tool development. The target users incorporate two counterparties, the lending protocols and lenders, both of whom can use the platform prototype to query various lending risk-related information. The project consists of two layers: the comprehension and refinement of machine learning models, and the web tool design and development. For the first layer, we will review and analyze current models, including credit scoring models in traditional lending businesses and user health estimation models in some top-ranking DeFi lending protocols. Evaluation metrics will be designed to compare the efficiency and accuracy of different models, and the model with desirable results will be selected and optimized to cater to the generalized DeFi lending risk assessment purpose more specifically.

In the second layer, the project is dedicated to building a website prototype, visualizing the machine learning results in a user-friendly way, and providing references on the risk levels of different accounts. Some add-on features like warning extensions will be appended after the basic functions are developed. The scope will be limited to the most active network first, whether Ethereum, Polygon, or Matic, and expanded to other networks if possible. The project is expected to offer a more resilient, scalable, and responsive solution to DeFi lending risk control, thus maintaining the stability and security of the DeFi market.