Paper: arXiv 2507.14160

Authors: Aaron Green, Zihan Nie, Hanzhen Qin, Oshani Seneviratne, Kristin P. Bennett

Abstract

Survival modeling predicts the time until an event occurs and is widely used in risk analysis; for example, it’s used in medicine to predict the survival of a patient based on censored data. There is a need for large-scale, realistic, and freely available datasets for benchmarking artificial intelligence (AI) survival models. In this paper, we derive a suite of 16 survival modeling tasks from publicly available transaction data generated by lending of cryptocurrencies in Decentralized Finance (DeFi). Each task was constructed using an automated pipeline based on choices of index and outcome events. For example, the model predicts the time from when a user borrows cryptocurrency coins (index event) until their first repayment (outcome event). We formulate a survival benchmark consisting of a suite of 16 survival-time prediction tasks (FinSurvival). We also automatically create 16 corresponding classification problems for each task by thresholding the survival time using the restricted mean survival time. With over 7.5 million records, FinSurvival provides a suite of realistic financial modeling tasks that will spur future AI survival modeling research. Our evaluation indicated that these are challenging tasks that are not well addressed by existing methods. FinSurvival enables the evaluation of AI survival models applicable to traditional finance, industry, medicine, and commerce, which is currently hindered by the lack of large public datasets. Our benchmark demonstrates how AI models could assess opportunities and risks in DeFi. In the future, the FinSurvival benchmark pipeline can be used to create new benchmarks by incorporating more DeFi transactions and protocols as the use of cryptocurrency grows.

Complexity vs Empirical Score

  • Math Complexity: 3.0/10
  • Empirical Rigor: 8.0/10
  • Quadrant: Street Traders — empirical and applied, lower math complexity

Why this score: The paper focuses on creating a large-scale benchmark dataset from DeFi transactions with minimal advanced mathematical theory, relying on standard survival analysis concepts and automated pipelines. It emphasizes practical implementation, data volume (7.5M records), and empirical evaluation of models on real financial data, making it highly ready for backtesting and application.

Research Flowchart

  flowchart TD
  A["Research Goal: Create Large-Scale<br>Financial Survival Benchmark"] --> B["Data Source:<br>Public DeFi Transaction Data"]
  B --> C["Methodology:<br>Automated Pipeline Construction"]
  C --> D["Core Output:<br>FinSurvival Suite<br>16 Survival Tasks<br>7.5M Records"]
  C --> E["Comparison Output:<br>16 Classification Tasks<br>via Thresholding"]
  D --> F["Evaluation:<br>Existing Methods<br>Struggle / Challenging"]
  E --> F
  F --> G["Outcome: Enabling AI<br>Survival Model Research"]

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper introduces FinSurvival, a benchmark suite of 16 large-scale survival modeling tasks derived from DeFi cryptocurrency lending transactions.

Source sentence
we derive a suite of 16 survival modeling tasks from publicly available transaction data generated by lending of cryptocurrencies in Decentralized Finance (DeFi).

Assumes

No explicit assumptions found in the text.

Data

  • FinSurvival · this paper crypto-onchain public
    Source sentence
    ... consists of 16 different time-to-event scenarios and combines for 7,698,497 records, averaging over 481,000 records per dataset.

Recognised sources: The Graph (subgraphs)

Costs

Not applicable The paper reports no trading or execution result.

Finds 3

  • Traditional models (XGBoost and AFT regression) outperformed deep learning methods in survival prediction, with XGBoost delivering the highest discriminative power. performance
    Source sentence
    In survival prediction, traditional models such as XGBoost and AFT regression outperformed deep learning methods, with XGBoost delivering the highest discriminative power.
  • For classification using RMST thresholds, logistic regression and Elastic Net achieved the best AUC scores, while deep‑learning models showed only moderate performance. predictive-accuracy
    Source sentence
    For classification tasks using RMST-based thresholds, basic logistic regression and Elastic Net led in AUC score, and deep‑learning methods demonstrated moderate performance.
  • Deep learning models performed poorly despite extensive tuning, indicating conventional approaches remain stronger in this complex financial setting. negative
    Source sentence
    ... we spent significant time tuning these methods, trying to achieve comparable results, and were not able to do so.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 25 related papers). Coverage is partial, so this is neither replication nor refutation.

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.

Cited References (64)

  1. Deep survival analysis (2016)
  2. Deep survival analysis: Nonparametrics and missingness (2018)
  3. A deep survival analysis method based on ranking (2019)
  4. Deep learning for survival analysis: a review (2024)
  5. A dynamic credit scoring model based on survival gradient boosting decision tree approach (2021)
  6. Probability of default for lifetime credit loss for IFRS 9 using machine learning competing risks survival analysis models (2024)
  7. Discrimination in mortgage lending: A literature review (2018)
  8. SurvSet: An open-source time-to-event dataset repository (2022) — link
and 56 more
  1. AAVE Protocol Whitepaper V2.0 (2020) — link
  2. AAVE Protocol Whitepaper (2020) — link
  3. Ethereum: A next-generation smart contract and decentralized application platform (2014) — link
  4. Characterizing Common Quarterly Behaviors In DeFi Lending Protocols (2023) — link
  5. DeFi Survival Analysis: Insights Into the Emerging Decentralized Financial Ecosystem (2024)
  6. DeFi survival analysis: Insights into risks and user behaviors (2022)
  7. DeepSurv: personalized treatment recommender system using a Cox proportional hazards deep neural network (2018)
  8. Neural Frailty Machine: Beyond proportional hazard assumption in neural survival regressions (2023)
  9. Survivalgan: Generating time-to-event data for survival analysis (2023)
  10. The SUPPORT prognostic model. Objective estimates of survival for seriously ill hospitalized adults. Study to understand prognoses and preferences for outcomes and risks of treatments. (1995) — link
  11. The genomic and transcriptomic architecture of 2,000 breast tumors reveals novel subgroups (2012) — link
  12. A 30-year perspective (1975-2005) into the changing landscape of patients hospitalized with initial acute myocardial infarction: Worcester Heart Attack Study (2009)
  13. The Urokinase System of Plasminogen Activation and Prognosis in 2780 Breast Cancer Patients1 (2000)
  14. General Semiparametric Shared Frailty Model: Estimation and Simulation with frailtySurv (2018) — link
  15. Forecasting Bankruptcy More Accurately: A Simple Hazard Model (2001)
  16. On Cox Processes and Credit Risky Securities (1998) — link
  17. Frailty Correlated Default (2009) — link
  18. Nonparametric Estimation from Incomplete Observations (1958)
  19. survminer: Drawing Survival Curves using ‘ggplot2’ (2021) — link
  20. Regression Models and Life-Tables (1972)
  21. R: A Language and Environment for Statistical Computing (2022) — link
  22. Random survival forests (2008) — link
  23. A Package for Survival Analysis in R (2022) — link
  24. DeepHit: A Deep Learning Approach to Survival Analysis With Competing Risks (2018) — link
  25. High-Dimensional Survival Analysis: Methods and Applications (2023) — link
  26. Restricted mean survival time: an alternative to the hazard ratio for the design and analysis of randomized trials (2014) — link
  27. Empirical choice of time window for restricted mean survival: Statistical approaches and applications (2021) — link
  28. Handling imbalanced datasets: A review (2006)
  29. Regression shrinkage and selection via the lasso (1996)
  30. Regularization and variable selection via the elastic net (2005)
  31. A decision-theoretic generalization of on-line learning and an application to boosting (1997)
  32. An empirical study of the naive Bayes classifier (2001)
  33. Classification and Regression Trees (1984) — link
  34. Random forests (2001)
  35. SMOTE: synthetic minority over-sampling technique (2002)
  36. Nearest neighbor pattern classification (1967)
  37. Evaluation: from precision, recall and F-measure to ROC, informedness, markedness and correlation (2011)
  38. Statistical comparisons of classifiers over multiple data sets (2006)
  39. The balanced accuracy and its posterior distribution (2010)
  40. Multiple comparisons using rank sums (1964)
  41. A flexible parametric accelerated failure time model and the extension to time-dependent acceleration factors (2022) — link
  42. SurvBenchmark: comprehensive benchmarking study of survival analysis methods using both omics data and clinical data (2022) — link
  43. Cross-Platform Omics Prediction procedure: a game changer for implementing precision medicine in patients with stage-III melanoma (2020) — link
  44. curatedOvarianData : clinically annotated data for the ovarian cancer transcriptome (2013)
  45. Greedy function approximation: A gradient boosting machine. (2001) — link
  46. XGBoost: A Scalable Tree Boosting System (2016) — link
  47. fastDummies: Fast Creation of Dummy (Binary) Columns and Rows from Categorical Variables (2025) — link
  48. R: A Language and Environment for Statistical Computing (2024) — link
  49. gbm: Generalized Boosted Regression Models (2024) — link
  50. xgboost: Extreme Gradient Boosting (2024) — link
  51. survivalmodels: Models for Survival Analysis (2024) — link
  52. The use of the area under the ROC curve in the evaluation of machine learning algorithms (1997) — link
  53. rpart: Recursive Partitioning and Regression Trees (2023) — link
  54. Regularization Paths for Generalized Linear Models via Coordinate Descent (2010) — link
  55. Elastic Net Regularization Paths for All Generalized Linear Models (2023) — link
  56. reticulate: Interface to ‘Python’ (2024) — link

0 of 64 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.