A Graph Correlated Anomaly Detection with Fuzzy Model for Distributed Wireless Sensor Networks

Wireless sensor networks have limited power for processing data, storage, and communication. Due to power shortages and anonymous attacks, sensor nodes may produce faulty or anomaly data which affects the accuracy of the entire system. Effective anomaly detection is essential to make an accurate prediction of the result. Moreover, clustering-based anomaly detection reduces energy consumption by avoiding individual sensory data reporting to the base station. The proposed methodology consists of two phases: Correlated graph clustering, and anomaly detection using a Fuzzy model. In the first phase, the spatial correlation of the sensor readings is used to generate a graph, partitioned into clusters. The intra-cluster and inter-cluster temporal correlations are analyzed to refine the optimized cluster structure. Finally, a fuzzy Mamdani model is used to classify the clusters as either normal or anomalous based on their membership values. The proposed approach leverages both spatial and temporal correlation between sensor measurements to form optimized clusters that are more effective for anomaly detection. The Experiments performed on a real-world dataset of WSNs indicate the efficacy of the proposed methodology, which shows significant improvement over traditional anomaly detection methods the electronic file of your paper will be formatted further for final publication.