Next-generation AI solutions for transaction security in digital finance

Cybersecurity threats in financial transactions have intensified with the growing adoption of digital financial platforms, necessitating advanced, scalable solutions. This study evaluates the effectiveness of LightGBM, Attention-Based Neural Networks, and CatBoost models in enhancing the security of financial systems. LightGBM was employed to detect fraud by uncovering complex patterns in transactional data, utilizing both numerical and categorical features. Attention mechanisms were incorporated to improve model accuracy by prioritizing relevant features for fraud detection. Sequential transaction data was analyzed using CatBoost, a gradient boosting algorithm optimized for categorical features, which performed well in identifying fraudulent patterns in imbalanced datasets. The dependent variables measured were Detection Accuracy (DA), False Positive Rate (FPR), and Privacy Preservation Index (PPI). Results showed that LightGBM achieved the highest DA (92%) in detecting complex fraud patterns, while CatBoost excelled in handling sequential transaction data with an FPR of 2%. Attention mechanisms demonstrated a PPI of 96%, ensuring compliance with privacy regulations like GDPR. Analysis of variance indicated significant improvements across all variables (p-value ≤ 0.05). The integrated use of LightGBM, Attention Mechanisms, and CatBoost provides a comprehensive approach to addressing evolving financial cybersecurity threats, offering a scalable, privacy-compliant solution that outperforms traditional methods.