PAANN-ProAct Autoencoder Neural Network Model for Energy Efficient and Robust Anomaly Detection in WSN

Wireless Sensor Networks (WSNs) are extensively employed in environmental monitoring, healthcare, industrial automation, military surveillance, and security applications. However, the limited energy, computational capacity, and communication resources of sensor nodes make efficient anomaly detection a challenging task. Malicious attacks, sensor faults, communication failures, and abnormal measurements can further degrade network reliability and reduce its operational lifetime. To overcome these challenges, this research proposes a ProAct Autoencoder Neural Network (PAANN) for energy-efficient and robust anomaly detection in WSNs. The proposed approach employs a cluster-based architecture in which sensor nodes perform local data processing and anomaly detection, while cluster heads aggregate the relevant information and forward significant anomaly events to the base station. The PAANN model utilizes an autoencoder to learn normal sensor-data patterns and detect abnormal observations based on reconstruction errors. Spatial correlation among neighboring sensor nodes is incorporated to improve anomaly identification and localization. An erroneous-decision abstainer mechanism is also introduced to reduce false-positive alerts and unnecessary computational operations. Furthermore, the Ascendable Energy Coherent Array-based Anomaly Detection (AECAAD) mechanism is incorporated to improve energy efficiency and identify potential Denial-of-Service (DoS) attacks from abnormal communication patterns. Huffman and Arithmetic Coding techniques are employed to compress data before transmission, thereby reducing communication overhead, bandwidth consumption, and energy expenditure. The proposed framework is designed to provide accurate and timely anomaly detection while extending network lifetime and improving the reliability, security, and robustness of WSNs.