A wireless sensor networks anomaly detection method based on the improved isolation forest
Reenu Batra, Manish Kumar, Deepak Kaushik, Megha Sharma, Aarti Sangwan · 2025
Sensors capture and store vast volumes of sensory data as a result of the ongoing development of technologies like the Internet of Things (IoT) and cloud computing, enabling real- time recording and awareness of the environment. The open nature of wireless sensor networks (WSN) makes network assault or infiltration feasible, and security threats during information transfer are significant. As a result, efficient anomaly detection is essential for IoT systems to maintain system security. The original Isolation Forest technique is a linear time complex anomaly detection system that performs better on perceptual data. However, there are drawbacks as well, including high unpredictability, poor generalisation, and insufficient stability. In order to solve the issues, this research suggests an isolation forest-based data anomaly detection approach for wireless sensor networks called BS-iForest (BoxPlot Sampled iForest). This approach initially trains and builds trees using the subset of data that has been filtered by the box graph. Following that, isolated trees from the training set that have greater accuracy are chosen to create a base forest anomaly detector. The basic forest anomaly detector will then assess data outliers for the following period using anomaly detection. These tests were conducted using datasets gathered from sensors placed in a university data centre as well as the BreastWisconsin (BreastW) dataset, which displays how well the variation of the Isolation Forest algorithm performs. The area under the curve (AUC) rose by 1.5% and 7.7% when compared to the conventional isolation forest, demonstrating that the suggested technique outperforms the two datasets we used.