FinSurvival Challenge 2025
收藏资源简介:
FinSurvival 2025数据集是由伦斯勒理工学院等机构联合构建的Web3时序分析基准,包含来自Aave v3协议的2180万笔DeFi交易记录。该数据集通过16项生存预测任务建模用户行为跃迁,完整记录了链上借贷行为与后续结果(如清算或还款)的时间间隔,具有右截尾和时序漂移特性。数据经过严格的时序分割和审查处理,采用固定评估管道确保可复现性,为研究开放式金融系统中的用户留存、风险累积等时序挑战提供了高保真实验场。
The FinSurvival 2025 dataset is a Web3 time-series analysis benchmark jointly developed by Rensselaer Polytechnic Institute and other institutions, containing 21.8 million DeFi transaction records from the Aave v3 protocol. This dataset models user behavior transitions through 16 survival prediction tasks, fully recording the time intervals between on-chain lending behaviors and their subsequent outcomes such as liquidation or repayment, and exhibits right-censored and temporal drift characteristics. The data has undergone rigorous temporal splitting and curation processing, and a fixed evaluation pipeline is adopted to ensure reproducibility, providing a high-fidelity experimental testbed for researching temporal challenges in open financial systems including user retention and risk accumulation.



