CADM: An LSTM-Based Model for Detecting Creative Accounting in Time-Series Data from Saudi-Listed Companies

Studies on Saudi accounting practices have identified evidence of creative accounting in the financial statements of listed companies.Despite the application of various fraud detection methods, identifying legal but misleading manipulations remains challenging.This paper extends the Creative Accounting Detection Model (CADM), an LSTM-based model originally proposed by Bineid et al. (2023, 2024) for detecting creative accounting.Two versions, (CADM1) and (CADM2), were trained on two simulated datasets with different bases, achieving 100% and 95% accuracy, respectively.Testing on the energy sector (2019-2023), CADM1 identified one company as engaging in creative accounting, while CADM2 classified all companies as non-creative with greater confidence stability.The findings establish CADM as a robust, scalable solution for the early detection of financial manipulation.By combining predictive strength with explainability, CADM can be employed to advance current approaches to forensic accounting and risk analytics, offering valuable insights to regulators, auditors, and decision-makers.