Kevin Fulcer et al. explore the use of unsupervised learning techniques to enhance audit analytics by identifying anomalies as outliers in financial datasets in their article Application of Outlier Detection Methods in Audit Analytics. These methods enable auditors to uncover potentially high-risk transactions without the need for labeled outcomes, offering a scalable and efficient solution to address the complexities of modern auditing.
The paper introduces a novel two-stage framework for applying outlier detection methods in audit processes. In the initial stage, the framework evaluates the credibility of various algorithms by testing their alignment with auditors’ judgment on a small set of known outliers. Algorithms demonstrating high alignment are deemed credible and are subsequently used in the second stage to detect previously unrecognized anomalies within the dataset.
A critical contribution of the study is the focus on mitigating algorithm selection bias. Unlike traditional methods relying on one or two algorithms, the proposed framework integrates multiple credible algorithms and aggregates their outlier scores to enhance robustness and accuracy. This ensemble approach reduces the risk of missing subtle anomalies that may be overlooked by individual algorithms.
The framework was validated through case studies using real-world datasets, including procurement and revenue subledgers. The results demonstrated that the framework could successfully identify high-risk transactions that traditional methods might miss, such as transactions with missing documentation or non-compliance with business rules. The study also found that auditors considered stringent algorithm selection methods to yield more relevant and actionable results.
Despite its innovations, the paper acknowledges certain limitations. For instance, the complexity of some outlier detection methods may hinder auditors’ ability to interpret results effectively. Additionally, the reliance on a small set of known outliers for algorithm validation introduces subjectivity. The authors propose further research into optimizing outlier selection processes and improving algorithm interpretability.
This research highlights the potential of integrating advanced data analytics with professional judgment to improve the audit process. For a comprehensive understanding, the article is available in Accounting Horizons and can be accessed here.