Publication

Machine Learning in Audit of Operations: Predicting the Error Rate of European Funds

View document

Bibliographic Details
Summary:In the field of auditing of operations, to assess error rates for Book Values (BV) in a European funds population, Monetary Unit Sampling (MUS), a widely used sampling methodology, enables auditors to extrapolate about the population being materially misstated. The thesis delved into the potential of incorporating supervised Machine Learning (ML) regression models to innovate how error rates are identified in European funds, given the limited exploration of ML’s effectiveness in auditing. Likewise, the study focused on showcasing the capability of predicted error rates align closely with the true misstatement values operations, previously selected through MUS sampling method. A pure model-based conceptual framework approach was developed introducing supervised ML regression models, including tree-based algorithms, such as Decision Trees, eXtreme Gradient Boosting, Categorical Boosting and Random Forest Quantile, and the Multilayer Perceptron algorithm. Additionally, Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) metrics, data visualization tools and Repeated Stratified Cross-Validation were used for models’ assessment. Furthermore, data samples were provided by the IGF entity, from training, validation and testing purposes. Results exhibited the proficiency of ML algorithms in predicting specific pattern performances although generalization across error rates remained a challenge. This advancement sets the stage for moving beyond standard the standard MUS method towards Artificial Intelligence (AI) applications in audit of operations.
Main Authors:Silva, Sara Filipa Santos Pereira Fernandes da
Subject:Operational Audits Monetary Unit Sampling Machine Learning Supervised Regression Algorithms SDG 16 - Peace, justice and strong institutions
Year:2024
Country:Portugal
Document type:master thesis
Access type:open access
Associated institution:Universidade Nova de Lisboa
Language:English
Origin:Repositório Institucional da UNL
Description
Summary:In the field of auditing of operations, to assess error rates for Book Values (BV) in a European funds population, Monetary Unit Sampling (MUS), a widely used sampling methodology, enables auditors to extrapolate about the population being materially misstated. The thesis delved into the potential of incorporating supervised Machine Learning (ML) regression models to innovate how error rates are identified in European funds, given the limited exploration of ML’s effectiveness in auditing. Likewise, the study focused on showcasing the capability of predicted error rates align closely with the true misstatement values operations, previously selected through MUS sampling method. A pure model-based conceptual framework approach was developed introducing supervised ML regression models, including tree-based algorithms, such as Decision Trees, eXtreme Gradient Boosting, Categorical Boosting and Random Forest Quantile, and the Multilayer Perceptron algorithm. Additionally, Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) metrics, data visualization tools and Repeated Stratified Cross-Validation were used for models’ assessment. Furthermore, data samples were provided by the IGF entity, from training, validation and testing purposes. Results exhibited the proficiency of ML algorithms in predicting specific pattern performances although generalization across error rates remained a challenge. This advancement sets the stage for moving beyond standard the standard MUS method towards Artificial Intelligence (AI) applications in audit of operations.