A Multi-Tier Edge–Fog Intelligent Learning Framework for Detecting Financial Anomalies in Smart Cities

The rapid digitization of urban infrastructure has reshaped smart cities into highly interconnected ecosystems where financial transactions are continuously generated at the network’s periphery. In such settings, ensuring timely detection of financial anomalies is critically dependent on distributed, low-latency intelligence. This paper presents an Edge–Fog-centric hybrid anomaly detection framework tailored for smart city financial infrastructure. The proposed architecture leverages a multi-tier collaborative model, wherein edge devices—such as point-of-sale terminals, ATMs, and mobile sensors—are equipped with real-time sensing, computing, and classification capabilities to grade financial transactions on a scale of 1 to 7. At the edge layer, unsupervised autoencoders, LightGBM regressors, and Isolation Forests operate locally to assign grades, which are refined through peer-to-peer collaboration between proximate edge nodes for contextual consistency. Transactions with higher anomaly grades are escalated to the fog layer, where a supervised One-Class SVM performs deeper analysis using device-level metadata (IP, MAC) to reduce false negatives. This edge–fog synergy enables decentralized yet coordinated detection that meets the stringent latency and scalability requirements of smart cities. Comparative evaluation on real-world financial data demonstrates the superiority of the proposed architecture, achieving accuracy: 98.82%, sensitivity: 91.30%, specificity: 98.79%, and reducing the false negative rate to 8.70%, outperforming standalone models such as autoencoders (FNR: 50.44%), Isolation Forests (FNR: 35.50%), and SVMs (FNR: 56.38%). These results highlight the efficacy of edge–fog collaborative intelligence in delivering robust, scalable, and context-aware financial anomaly detection in next-generation smart cities.