Reading Between the Lines - Incorporating Text Mining and Machine Learning in Financial Fraud Detection

Aug 5, 2026

Upload by Ophelia

Despite rigorous oversight in the Indonesian capital market, manipulation of financial reports remains persistent. This study evaluates machine learning models that combine linguistic features and financial ratios to detect deception and financial statement manipulation. Using publicly listed Indonesian companies as samples, the research validates the predictive power of the Beneish M-Score, identifies negative language patterns in fraudulent reports, and shows Gradient Boosting outperforms in anomaly detection across financial and textual signals. The framework adapts to Indonesian-language annual reports, addressing gaps in linguistic-based fraud detection literature.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages21

Reading Time53 min

AccessFree to download

Introduction

Financial statement fraud and its consequences

Challenges in detection and need for advanced methods

Beneish M-Score and NLP-based detection

Research gap and study contribution

Asia Pacific Fraud Journal E-ISSN: 2502-695X, ISSN: 2502-8731 Volume 10, Issue 1 (January-June) 2025 Asia Pacific Fraud Journal now has been accredited “SINTA 3” by Ministry of Research and Technology of The Republic of Indonesia (No SK. 225/E/KPT/2022) . Available online at: [http://apfjournal.or.id/index.php/apf](http://apfjournal.or.id/index.php/apf) Reading Between the Lines: Incorporating Text Mining and Machine Learning in Financial Fraud Detection 􀀍 Agung Septia Wibowo & Iis Istianah Universitas Gadjah Mada, Indonesia ARTICLE INFORMATION Article History: Received November 13, 2024 Revised May 20, 2025 Accepted June 10, 2025 DOI: 10.21532/apfjournal.v10i1.382 This is an open access article under the CC-BY-SA License ABTRACT Notwithstanding rigorous oversight in the Indonesian capital market, the manipulation of financial reports continues to occur. This study examines the potential for employing machine learning (ML) models, which utilize linguistic features and financial ratios, in effectively detecting deception or manipulation. Drawing upon publicly listed Indonesian companies as the samples, this research validates the predictive capabilities of the Beneish M-Score, confirms the occurrence of negative language in fraudulent reports, and demonstrates the superiority of the Gradient Boosting ML model in identifying anomalies within financial and textual data. The study distinctively adapts to Indonesian-language annual reports, thereby addressing a gap in the linguistic-based fraud detection literature. These findings not only advance our comprehension of how linguistic features and financial ratios provide practical tools for fraud detection, thereby preparing the academic and professional community in this domain. Keywords: Financial Statement Fraud, Text Mining, Machine Learning (ML), Natural Language Processing (NLP), Financial Ratios. How to Cite: Wibowo A. S., & Istianah I. (2025) . Reading Between the Lines: Incorporating Text Mining and Machine Learning in Financial Fraud Detection. Asia Pacific Fraud Journal, 10(1), 73-93. [http://doi](http://doi). org/10.21532/apfjournal.v10i1.382 . 􀀍 Corresponding author : Association of Certified Fraud Examiners (ACFE) Email: [agungseptiawibowo@gmail.com](agungseptiawibowo@gmail.com) Indonesia Chapter *The 2nd Best Paper of National Call for Paper ACFE Indonesia Chapter 2024 Page. 73-93 74| A. S. Wibowo & I. Istianah, Reading Between the Lines: Incorporating Text Mining 1. INTRODUCTION Financial statement fraud (FSF) has farreaching consequences. It impacts stakeholders both financially and emotionally, affecting entire communities. When financial reports are manipulated, the resulting information becomes inaccurate, leading to misguided decisions. The damage can be severe, ranging from lost of public trust to corporate bankruptcy. Research by Dechow et al. (2010) shows that manipulating financial statements often reduces share value and increases market volatility, ultimately harming shareholders and financial stability. Aghghaleh et al.(2016) found that financial statement fraud results in significant financial losses, with organizations losing an average of 6% of their annual revenue. High-profile cases-including Enron, WorldCom, and Satyam-caused losses exceeding US$20 billion, underscoring the urgent need for preventive measures. Hogan et al (2008) highlighted those 30 financial scandalsled to a loss of more than $900 billion in market capitalization. Proactive steps, such as independent audits, strict internal controls, and whistleblower hotlines, are crucial. According to “Occupational Fraud 2024: A Report to the Nations,” Financial Statement Fraud has a median loss of $766,000 per case, making it the least common but most expensive category of fraud (ACFE, 2024) . FSF is often concealed from the public and auditors, making prompt detection challenging (Aghghaleh et al., 2016) . The complexity of manipulating financial

FAQs

What problem does the study address in the Indonesian capital market?

The study addresses continued manipulation of financial reports despite regulatory oversight, focusing on how to detect deception in financial statement fraud.

Which methods does the study combine for fraud detection?

It integrates machine learning models with linguistic features extracted from text and financial ratios, validating the Beneish M-Score and using NLP-derived language signals.

Which machine learning approach performs best in detecting anomalies?

The Gradient Boosting machine learning model demonstrates superior performance in identifying anomalies using both financial and textual data.

Aug 5, 2026

Upload by Ophelia