fraud detection

both academic and auditing firms have been searching for ways to detect corporate fraud. The main objective of this study was to present a model to detect financial reporting fraud by companies listed on Tehran Stock Exchange (TSE) using genetic algorithm. For this purpose, consistent with theoretical foundations, 21 variables were selected to predict fraud in financial reporting that finally, using statistical tests, 9 variables including SALE/EMP, RECT/SALE, LT/CEQ, INVT/SALE, SALE/TA, NI/CEQ, NI/SALE, LT/XINT, and AT/LT were selected as the potential financial reporting fraud indexes. Then, using genetic algorithm, the final model of fraud detection in financial reporting was presented. The statistical population of this study included 66 companies including 33 fraudulent and 33 non-fraudulent companies from 2011 to 2016. The results showed that the presented model with the accuracy of 91.5% can detect fraudulent companies. These findings extend financial statement fraud research and can be used by practitioners and regulators to improve fraud risk models.

Relying on the Area Under the Curve (AUC) measure, we compare the performance of the Logit regression model and the LightGBM algorithm. Despite these methods being common in the literature, our study emphasizes the role of statistical inference to evaluate and compare the results comprehensively. We use the training set of the Vesta (2018) dataset, provided by Vesta—a global fraud prevention company headquartered in the United States specializing in payment solutions and risk management. Originally released as part of a Kaggle competition focused on credit card fraud detection, this dataset comprises diverse transaction records, representing a rich source for exploring advanced fraud detection methods. Our analysis reveals that while the LightGBM algorithm generally yields higher predictive accuracy, the differences between the calculated AUCs of the two methods are not statistically significant. This underscores the importance of using inferential techniques to validate model performance differences in fraud detection.

Internal Financial Control (IFC) is a critical component of corporate governance, ensuring the accuracy, reliability, and compliance of financial reporting. Traditional IFC systems rely on manual audits, centralized databases, and rule-based checks, which are often inefficient, prone to human error, and vulnerable to fraud. The integration of Blockchain Technology and Machine Learning (ML) has introduced transformative improvements in Internal Financial Control (IFC) systems. This paper explores how Blockchain and machine learning (ML) technologies can strengthen internal financial controls (IFC). By addressing limitations in traditional systems, these technologies introduce transparency, automation, and predictive capability, fostering enhanced compliance and reduced risk. The integration of these technologies offers a paradigm shift for governance, risk management, and auditing practices, enhances fraud detection and regulatory compliance, while addressing challenges such as scalability and data privacy. Through a synthesis of academic literature and industry case studies, Blockchain ensures immutable transaction records, while ML enables predictive anomaly detection. Blockchain and ML are transforming internal financial control by enhancing security, automation, and predictive capabilities. There are still challenges in overcoming scalability, interpretability, Hybrid Blockchain-ML frameworks, and regulatory challenges for widespread adoption.

The rapid growth of digital payment systems has heightened the need for accurate and scalable methods to detect credit card fraud. This study evaluates a range of machine learning and deep learning algorithms, including Logistic Regression, Decision Tree, Random Forest, K-Nearest Neighbors (KNN), XGBoost, Convolutional Neural Networks (CNN), Baseline MLP (Multi-Layer Perceptron), and Long Short-Term Memory (LSTM), to identify effective approaches for detecting fraudulent transactions. Based on comparative analysis, Random Forest and LSTM achieved the strongest individual performance, with accuracies exceeding 96%. Building on these findings, a stacking ensemble model was constructed by integrating Random Forest and LSTM as base learners and Logistic Regression as the meta-classifier. The framework incorporates Convolutional Autoencoder (CAE) for feature extraction and Random Undersampling (RUS) with three resampling ratios (1:1, 1:5, and 1:10) to address class imbalance. Experimental results show that the ensemble model provides improved predictive accuracy compared with individual algorithms, achieving an accuracy of 99.98%, precision of 99.86%, and recall of 99.89% under a 1:10 resampling ratio. Rather than proposing a new algorithmic architecture, this study contributes a systematic and unified evaluation of widely used ML and DL approaches and demonstrates the effectiveness of integrating CAE, RUS, and a Random Forest–LSTM stacking ensemble in enhancing fraud detection performance.