Intelligent FinTech Risk Management Using Machine Learning-Based Anomaly Detection and Predictive Analytics

Abstract

The rapid expansion of digital financial services has increased transaction volumes and introduced increasingly complex financial-risk patterns, creating a need for reliable computational approaches to detect fraudulent activity. Machine-learning techniques can support FinTech risk management by identifying relationships and irregularities within transaction data that may not be readily recognized through conventional monitoring methods. This study aimed to characterize fraudulent financial-payment activity across transaction categories and assess the performance of supervised and unsupervised machine-learning methods for financial-risk identification. The BankSim financial-payment transaction dataset was analyzed using descriptive statistical techniques and machine-learning methods. Transaction distributions, monetary characteristics, and category-specific fraud patterns were examined, followed by the implementation of Logistic Regression, Random Forest, and Isolation Forest. An 80:20 stratified training and testing framework was applied, and model performance was evaluated using accuracy, precision, recall, F1-score, Area Under the Receiver Operating Characteristic Curve (AUROC), and Area Under the Precision-Recall Curve (AUPRC). The analysis identified substantial differences in fraud occurrence among transaction categories and demonstrated distinct performance profiles across the evaluated approaches. The findings indicate that supervised classification and anomaly detection provide complementary perspectives for financial fraud identification. Integrating transaction-level characteristics with machine-learning analysis can support structured risk assessment, improve the identification of potentially suspicious activity, and contribute to more responsive FinTech monitoring frameworks. Future investigations should examine adaptive and advanced machine-learning approaches using broader financial transaction environments.