A Multivariate Outlier Detection Algorithm for Wireless Sensor Networks
Chafiq Titouna, Farid Naït‐Abdesselam, Ashfaq Khokhar · 2019
In wireless sensor networks, an outlier detection algorithm removes from sensed data any possible error, redundancy and malicious injection of fake data. Therefore, such algorithms improve considerably the reliability and the accuracy of the collected data and also reduce the overall energy consumption in the context of large-scale and dense networks. Yet, the proposed algorithms generate huge communication costs due to information exchanges among neighbors and cannot detect outliers for more than one data type. In this paper, we propose a new outlier detection algorithm that is capable of detecting outliers even in the presence of multiple types of data and that does not require any information about the neighborhood. The proposed approach relies on a set of classifiers implemented in each node of a wireless sensor network. The sensed data is therefore classified in either outlier or normal data in a distributed manner. The extensive simulations of the proposed algorithm confirm its outperforming characteristics in terms of detection accuracy, false alarm rate and energy consumption.