Fraud-Resilient Banking Through Hybrid Artificial Intelligence and Zero-Knowledge Proofs

Abstract

Detecting fraud in digital banking is a recognized challenge given the increasing sophistication of perpetrators as well as the limitations of traditional security models. Rules-based systems produce interpretability but cannot be adapted to emerging fraud threats. Advanced machine learning models require complex systems for training and serving, which are often not practical in lightweight and real-time environments. In this work, a Java-based Hybrid Framework for Fraud-Resilient Banking Systems is built that combines rule-based compliance, lightweight AI modeling, anomaly detection, and the application of cryptographic concept of Zero Knowledge Proof (ZKP) authentication in a decision layer. A synthetic dataset of 10,000 banking transactions representing realistic imbalances has been developed, with approximately 0.6% flagging transactions as fraudulent. The rules for interpretable transparency are applied in the event of high-value transactions or merchant transactions that are shown to be suspicious, the fraud detection is tackled through a logistic regression classifier to apply a probabilistic approach to fraud detection and z-scores have been used to identify anomalous outliers from an expected normal distribution as fraud. The framework included Schnorr’s ZKP protocol to authenticate the user without disclosing the secret credential. The consolidated scoring system incorporates the outputs of rules, AI probabilities, anomalies, and ZKP verification for sorting transactions into High, Medium, and Low risk. The experimental results on the Java implementation shows an achievable ROC AUC of 0.984. The system produces a balanced risk distribution, for 1.8% of transactions classified as high risk, 20% medium risk and 78% low risk. This research suggests that a lightweight Java-based fraud detection system can be made efficient, interpretable, and cryptographically augmented, and thus usable in practice for a banking platform where performance and security are key.