Semisupervised Rule Induction for Multiperiod Audit Analytics

ABSTRACT

The aim of this study is to introduce a design science research artifact that will integrate semisupervised learning into rule induction and exception selection in audit analytics to exploit the structural information in the availability of unlabeled transactions that remain uninvestigated because of budget limitations. By leveraging both labeled and unlabeled data, this method is expected to enhance rule induction accuracy, improve exception selection, and strengthen the feedback loop between detection and rule induction in multiperiod audit settings. Using simulation experiments on error-seeded payroll data, we evaluate semisupervised rule induction models against the existing models of current literature and demonstrate the ability of semisupervision to detect a greater proportion of seeded misstatements under fixed audit budgets. We believe that the expected results will highlight the value of semisupervised learning in detecting errors and deriving more effective rules, optimizing audit resource allocation, and advancing the design of intelligent audit analytics systems.

Data Availability: Although we do not use any data with personal identifiable information, we do not have any permission from the data provider to share the original dataset publicly. However, we can share the source code and output data from our simulations that do not hold any part of the original dataset.