Abstract
The integration of self-adjusting AI into financial technology marks a breakthrough in detecting credit card fraud and abnormal transactions. This chapter presents a flexible framework using ML techniques, XGBoost, Isolation Forest, Federated Learning, and SHAP, to automate fraud detection with high precision. Transactions are first assessed via XGBoost, then anomalies (e.g., small purchases by younger users) trigger Isolation Forest. These are added to the labeled dataset, enabling real-time model retraining. Threshold tuning enhances sensitivity, while SHAP ensures interpretability. A simulated federated learning setup updates models across institutions without compromising privacy. The visual feedback loop illustrates continuous learning and deployment. This adaptive system boosts fraud detection accuracy, reduces false positives, and strengthens institutional agility, driving innovation, trust, and scalable value in FinTech.