۱- Introduction

۲- Related work

۳- Distributed outlier detection scheme

۴- Performance evaluation

۵- Conclusion

References

Abstract

In many wireless sensor network (WSN) applications, where a plethora of nodes are deployed to sense physical phenomena, erroneous measurements could be generated mainly due to the presence of harsh environments and/or to the depletion of a sensor’s battery. The measurements that significantly deviate from a normal behavior of sensed data are considered as outliers. To address the problem of detecting these outliers in wireless sensor networks, we propose a new algorithm, called Distributed Outlier Detection Scheme (DODS), in which multiple sensed data types are considered and where outliers are detected locally by each node, using a set of classifiers, so that neither information about neighbors is needed to be known by other nodes nor a communication is required among them. These characteristics allow the proposed scheme to be scalable and efficient in terms of both energy consumption and communication cost. The functionalities of the proposed scheme have been validated through extensive simulations using real sensed data obtained from Intel-Berkeley Research Lab. The obtained results demonstrate the efficiency of the proposed scheme in comparison to the surveyed algorithms.

Introduction

The advances in the fields of transistors and semiconductor devices have led to the deployment of wireless sensor networks (WSNs). A wireless sensor network (WSN) is a self-organized network that consists of a large number of low-cost and low-powered sensor devices, which can be deployed in a field, in the air, in vehicles, on bodies, underwater, and inside buildings. These small sensing devices can cooperatively monitor real world physical or environmental conditions, such as temperature, pollution, pressure, light, voltage, humidity and motion. They are also considered as particular networks which are widely used in commercial and industrial areas, for example, transportation tracking, environmental and habitat monitoring, healthcare, etc. Moreover, in a military applications, WSNs can be used for target tracking and battlefield surveillance. In many of these applications, the data sensed by nodes are often unreliable. The quality of the data is affected by multiple noises and errors, missing values, duplicated data, or inconsistent data [1], without forgetting the low performance of nodes in terms of energy, computational and memory capabilities. These issues generally lead into having the generated data unreliable and inaccurate. One of the most sources that influence the quality of sensed data are outliers. We can define outliers as those measurements that significantly deviate from the normal pattern of the sensed data [1]. It means that the sensed data should be in coherence with a pattern which represents the reality of the sensed data. Therefore, it is clear that outlier detection is a crucial task in WSNs as it improves the quality of data, the security of the system, and maximizes the lifetime of the network.