| Abstract: |
The wireless sensor networks (WSNs) play a significant role in the new monitoring and control systems and the performance of these systems is normally influenced by failures, bad intent attacks and the uncharacteristic actions of the nodes. The paper discusses machine learning-based WSNs anomaly detection algorithms to enhance the communication of such networks under both dynamic and limited resources. The quantitative version of an experimental method is used to compare the conventional threshold-based and statistical methods with the classical machine learning models (Support Vector Machine, Random Forest, and a new ML-based anomaly detection model). The criteria of performance evaluation are the standard detection measures that consist of accuracy, precision, recall, F1-score, false alarm rates, network lifetime and necessary key network measures that consist of the packet delivery ratio, end-to-end delay, and normal network lifetime. The proposed ML-based model has higher detection rate, significantly lower false positive and false negative rate and generally lower network work since it is able to provide more packets and increase the network lifespan with a low communication delay as demonstrated in experiments. These results confirm that adaptive machine learning-based anomaly detection offers a viable and robust solution in enhancing robustness and operational efficiency in Wireless Sensor Networks in any real world usage. |
