Scalable Anomaly Detection and Pattern Mining in Iot Aquaponics Systems Via Federated Learning

This paper proposes a federated learning framework for scalable and decentralized anomaly detection in IoT-enabled aquaponics systems. In this regard, the proposed framework discovers important sequential patterns across distributed datasets with much consideration of data privacy using PrefixSpan-based pattern mining. IoT sensors deployed across 11 aquaponics ponds collected over 170,000 readings per pond, monitoring critical parameters such as temperature, pH, turbidity, and fish growth metrics. The designed FL model has shown great congruity in water quality parameters with the growth of fishes in the ponds, showing values more than 0.9. Ponds 6 and 10 resulted in total counts of patterns as 1269 and 1339, respectively, which prove the capability of the system to scale. The presented FL framework reduced latency by 35 % compared to standard centralized approaches and thus enabled distributed and efficient anomaly detection. This approach addresses the challenges of data centralization and real-time monitoring; hence, it presents a robust solution in managing sustainable aquaponics.