SAILFISH OPTIMIZED ENSEMBLE LEARNING WITH FEDERATED LEARNING FOR ANOMALY DETECTION IN IOT EDGE NETWORKS

Abstract

Federated learning (FL) combines machine learning with privacypreserving distributed training across devices and edge nodes, making it ideal for IoT anomaly detection where central data collection is impractical. IoT adoption in India is rapidly expanding, The objective is to leverage federated learning and sailfish-optimized ensemble models to detect IoT anomalies efficiently at edge nodes, enhancing security and performance without centralizing sensitive data, thus ensuring privacy and scalability. In manual IoT security systems, administrators manually monitor network logs, set static thresholds for unusual behavior, inspect alerts by hand, and respond to anomalies after detection, often using rule-based tools and human intervention to investigate and remediate suspicious events. Manual systems are slow, error-prone, and unscalable; they struggle with high data volumes and evolving patterns, fail to detect subtle or novel anomalies, and cannot adapt in real time, leading to delayed responses and increased security risk. The motivation is to overcome the limitations of manual monitoring—such as delayed detection, lack of privacy protection, and inability to scale—by using federated learning with optimized ensemble models that enable real-time, distributed anomaly detection while preserving data privacy and efficiently handling heterogeneous IoT edge environments The proposed system integrates PAC (Probably Approximately Correct) frameworks, NBC (Naive Bayes Classifier), KNN (K-Nearest Neighbors), and CNN (Convolutional Neural Networks) within a federated learning architecture to detect anomalies in IoT edge computing environments. PAC provides theoretical guarantees on learning performance, NBC offers fast probabilistic classification for lightweight devices, and KNN aids in instance-based detection of abnormal patterns. CNNs extract deep spatial/temporal features from sensor data, improving detection accuracy. Sailfishoptimized ensemble techniques combine these models to balance precision, recall, and robustness. Federated learning trains models across edge nodes without sharing raw data, enhancing privacy, reducing communication overhead, and adapting to device heterogeneity for effective real-time IoT anomaly detection.