Abstract:
This research utilizes financial statement data from listed companies on the Stock Exchange of Thailand, spanning over six fiscal years, to construct an Artificial Neural Network (ANN) model. This model provides a high level of accuracy in fraud prediction, serving as a reliable tool for certified public accountants to conduct auditing tasks. It leverages electronic data processed through the M-Score to detect embellishments in financial statements. It trains the neural network to learn and accurately predict fraud in new, previously unseen data. The neural network is structured for deep learning with supervised learning techniques, employing feedforward and backpropagation algorithms with a sigmoid activation function and calibration. The research found that the artificial neural network has an overall accuracy of 94.02% and a post-calibration reliability of 80.41%, indicating that the neural network is efficient and reliable. It effectively identifies and highlights the feature importance used in its training process, extracting critical accounting entries such as sales, accounts receivable, inventory, purchases, and accounts payable. These are the accounting entries that auditors are authorized to utilize with Computer Assisted Audit Techniques (CAATs) during the audit process. Additionally, there is potential to develop these applications to work with the Internet of Things (IoT) in real-time.