VynFi/vynfi-journal-entries-1m
收藏资源简介:
VynFi Journal Entries — 1M (v2, v5.29 SOTA模式) 是一个合成总账(GL)数据集,专为会计和欺诈检测研究设计。该数据集包含100万行日记账条目,基于10家公司×12个月×10万行的多货币框架生成。它是v1版本的升级版,采用了SOTA-N结构保真度改进和集中浓度抽象(ConcentrationPipeline),在13个结构指标上更接近参考基线,行为保真度(基于Sajja 2026 P1-P4框架)的波动校正综合指标相比v1提升了43%。数据集用于表格分类、表格回归等任务,具有合成、会计、审计、SAP、金融和基准等标签,适用于机器学习训练和探索。
VynFi Journal Entries — 1M (v2, v5.29 SOTA mode) is a synthetic general ledger (GL) dataset designed for accounting and fraud detection research. It contains 1 million journal entry lines, generated from a 10-company × 12-month × hundred_k multi-currency scaffold. This release replaces v1 (v5.27) with SOTA-N structural-fidelity improvements and the central concentration abstraction (ConcentrationPipeline). 13 measured structural metrics moved toward the reference baseline, and behavioral fidelity (Sajja 2026 P1-P4 framework) improved by 43% in the vol-corrected composite compared to v1. The dataset is used for tabular classification, tabular regression, and other tasks, with tags including synthetic, accounting, fraud detection, audit, SAP, financial, and benchmark, suitable for machine learning training and exploration.




