AN ADVANCED AI-EMPOWERED FINTECH FRAMEWORK FOR CREDIT CARD FRAUD DETECTION IN ONLINE TRANSACTIONS VIA SPARROW SEARCH ALGORITHM–BASED ADAPTIVE OPTIMIZATION

Abstract

The rapid growth of online financial transactions has significantly increased the exposure of digital payment systems to sophisticated credit card fraud, posing serious challenges to financial institutions and consumers alike. The highly imbalanced nature of transactional datasets, evolving fraud patterns, and stringent real-time decision requirements often limit the effectiveness and stability of conventional machine learning and deep learning approaches. To address these challenges, this study proposes an advanced AI-empowered FinTech framework for credit card fraud detection in online transactions that integrates adaptive optimization based on the Sparrow Search Algorithm (SSA) to enhance detection performance, robustness, and operational feasibility. The proposed framework follows a structured end-to-end pipeline comprising secure data ingestion, preprocessing and feature normalization, imbalance-aware learning, SSA-driven hyperparameter and decision-threshold optimization, and comprehensive performance evaluation under cost-sensitive constraints. At the core of the framework, SSA is employed as a global metaheuristic optimizer to adaptively tune critical model parameters across multiple candidate learners, including interpretable classifiers, ensemble methods, and lightweight neural architectures. Unlike conventional grid or random search techniques, the SSA-based optimization dynamically balances global exploration and local exploitation, enabling efficient navigation of complex hyperparameter spaces while maintaining computational efficiency. The optimization objective explicitly incorporates cost-sensitive learning by penalizing false negatives more heavily than false positives, reflecting the asymmetric financial risk associated with undetected fraudulent transactions. This design ensures that the optimized models align with real-world fraud management priorities rather than purely accuracy-driven objectives. The framework further integrates robust evaluation protocols using stratified cross-validation and hold-out testing to assess generalization capability and stability. Performance is measured using fraud-specific metrics, including precision–recall area under the curve, recall, F1-score, Matthews correlation coefficient, balanced accuracy, and expected financial cost. Comparative analyses against non-optimized baselines demonstrate that SSA-optimized models achieve superior minority-class detection, reduced performance variance, and improved cost efficiency across varying fraud prevalence levels. In addition, the proposed architecture considers practical deployment aspects such as inference latency, model monitoring, and periodic re-optimization to address concept drift in evolving transaction streams. Overall, this study contributes a cost-aware, optimization-driven, and deployment-ready AI framework that strengthens the resilience of online payment systems. The proposed SSA-based adaptive optimization strategy offers a scalable and effective solution for enhancing credit card fraud detection performance in modern FinTech environments.