Developing Predictive Models for Detecting Financial Statement Fraud: A Machine Learning Approach

The objective of this study is to overcome the shortcomings of conventional ways to detect fraud in financial statement analysis, including rule-based and statistical methods, which frequently fail to identify intricate patterns suggestive of fraud. This research aims to improve the detection of financial statement fraud through the development of a machine learning-based predictive model, thereby enhancing the integrity of financial markets and mitigating significant economic losses. The study utilizes an extensive dataset comprising financial ratios, governance indicators, and company-specific attributes to train multiple machine learning models, namely Random Forest, XGBoost, and Support Vector Machines (SVM). Data preprocessing procedures, including scaling, addressing missing values, and class balancing via SMOTE, were implemented to guarantee dependable model training and validation. Results demonstrate that ensemble methods, specifically Random Forest and XGBoost, surpass conventional detection techniques by attaining enhanced accuracy, recall, and AUC-ROC scores. The analysis demonstrated that non-financial indicators, including audit fees and board independence, are crucial for detecting fraud, underscoring the importance of integrating governance-related data into fraud detection models. This study illustrates the benefits of machine learning models in detecting financial fraud and suggests a pragmatic framework for their application in auditing and regulatory environments. The study highlights the efficacy of ensemble methods, emphasizing their potential as data-driven, scalable solutions for improved corporate governance, financial oversight, and regulatory practices. Subsequent research could advance this work by incorporating alternative data sources, such as sentiment analysis, and expanding datasets to enhance model generalization.