Hybrid Machine Learning Model for Anomaly Detection in Unlabelled Data of Wireless Sensor Networks

A wide variety of sectors or fields are highly dependent on the data collected by the wireless sensor networks, so even a bit of unreliable or inaccurate data has the potential to cause a great harm to the society as a whole. Thus the urgency to flag the anomalous data and its source becomes vital. In this paper, a Hybrid Model of One-class SVM and Isolation forest (HMOI) has been proposed. The proposed model is a 'Classification + Classification' model which has two essential phases. The first phase addresses the issue of unlabelled data, which is most common in case of real-world dataset of wireless sensor networks and converts it into labelled data. The second phase is dedicated for the task of anomaly detection. It detects the anomalous data and flags the corresponding mote-id of the anomalous sensors in the network. The performance of the model has been tested and evaluated on the basis of evaluation metrics such as recall, precision, F-measure, accuracy, false alarm rate and AUCPR. The results show that the proposed model outperformed the existing models or techniques by achieving 99.92% recall, 99.87% precision, 99.89% f-measure, 99.78% accuracy, 0.08% false alarm rate and 99.99% AUCPR. The proposed model has been evaluated over the Intel Berkeley Research Lab (IBRL) dataset.