Publikation
Generating Synthetic Journal Entries for Audit Analytics: Method and Dataset
Jan Gronewald; Sebastian Stephan; Alexander Michael Rombach; Peter Fettke
In: Data in Brief, Vol. Online, Pages 1-19, Elsevier, 2026.
Zusammenfassung
This article presents a labeled synthetic general ledger dataset created for journal entry testing in audit analytics. Journal entry testing is a use case of financial audit that deals with the detection of erroneous or fraudulent entries in the general ledger. The dataset represents two fiscal years of transactions for a fictional company as recorded in an accounting system. The postings are generated to comply with the rules of double-entry bookkeeping, with each journal entry consisting of at least two posting lines. In addition to regular transactions, the dataset contains transactions that would be classified as problematic in the annual audit. The synthetic postings were generated using predefined transaction patterns and probabilistic distributions for transaction frequency and other attribute values. The dataset can be reused for the development and benchmarking of supervised, unsupervised, rule-based, and hybrid approaches to anomaly detection in accounting data. The generation workflow can also serve as a reusable blueprint for constructing synthetic accounting datasets in related contexts.