Multivariate Statistical Approach for Anomaly Detection and Lost Data Recovery in Wireless Sensor Networks

Data loss due to integrity attacks or malfunction constitutes a principal concern in wireless sensor networks (WSNs). The present paper introduces a novel data loss/modification detection and recovery scheme in this context. Both elements, detection and data recovery, rely on a multivariate statistical analysis approach that exploits spatial density, a common feature in network environments such as WSNs. To evaluate the proposal, we consider WSN scenarios based on temperature sensors, both simulated and real. Furthermore, we consider three different routing algorithms, showing the strong interplay among (a) the routing strategy, (b) the negative effect of data loss on the network performance, and (c) the data recovering capability of the approach. We also introduce a novel data arrangement method to exploit the spatial correlation among the sensors in a more efficient manner. In this data arrangement, we only consider the nearest nodes to a given affected sensor, improving the data recovery performance up to 99%. According to the results, the proposed mechanisms based on multivariate techniques improve the robustness of WSNs against data loss.