ETELMAD: Anomaly Detection Using Enhanced Transient Extreme Machine Learning System in Wireless Sensor Networks

Anomaly detection is a major task for ensuring security in WSNs, and they are sensitive to several attacks, which cause the node to break and generate faulty results. This proposed architecture introduces a method named Enhanced Transient Extreme Learning Machine Anomaly Detection to resolve such an issue. The detection of anomalies in the sensor data is recorded in three stages: data compression, prediction, and anomalous detection. The data collected from the network is pre-processed, and the duplicate values are eliminated from the dataset. Piecewise Aggregate Approximation is employed for the data compression process. This method can extract a low-dimensional set of features with less dimension and high accuracy. The reduction in dimensionality plays a major role in the WSN environment and attains less computation or training time. The second phase is prediction, done by an Extreme Learning Machine (ELM). The parameters of ELM are optimized by the meta-heuristic approach Enhanced Transient Search Arithmetic Optimization. Finally, the anomalous data is detected using the dynamic thresholding method. Dynamic thresholding is a process that generates a set of threshold values to differentiate the normal and abnormal sensed data. The PYTHON platform is used to simulate the proposed process. The achieved performance is compared over other models based on some measures to depict the efficacy of the proposed anomaly detection model. The overall accuracy achieved by this proposed architecture for the IBRL dataset is 97.4%, which is more efficient than other existing approaches.