A user-centric intelligent service for detecting financial statement fraud through optimized ensemble learning

Abstrak

As the complexity of the global economy increases, financial statement fraud continues to be a serious challenge in auditing practices. The process of conventional auditing practice often takes a long time and may not identify fraudulent activity early enough to prevent significant financial losses. While the advances in machine learning studies offer a promising way forward, the technical complexity of building and interpreting these models has slowed their adoption in real auditing practice. This study introduces an integrated approach that combines an Extreme Gradient Boosting (XGBoost) model, optimized with a Genetic Algorithm (GA), and a graphical user interface (GUI) designed to make the system accessible to end users. The approach is tested using data from public companies in Indonesia, serving as a case study. Results show that the model delivers high predictive performance, reaching an average accuracy of 97.91% as well as sensitivity of 97.13%. Further results showed that the developed GUI was intuitive based on usability testing, with an average task completion of 4.5 min and a satisfaction score of 4.2 out of 5. These findings suggest that the proposed system can serve as an effective as well as user-friendly tool for early fraud detection, with potential applications in a wide range of auditing contexts.

Identifier

• OpenAlex: https://openalex.org/W7160391829
• DOI: https://doi.org/10.1016/j.rineng.2026.110787