Multivariate Outlier Detection for Forest Fire Data Aggregation Accuracy

Abstract

Wireless sensor networks have been a very important means in forest monitoring applications. A clustered sensor network comprises a set of cluster members and one cluster head. The cluster members are normally located close to each other, with overlaps among their sensing coverage within the cluster. The cluster members concurrently detect the same event to send to the Cluster Head node. This is where data aggregation is deployed to remove redundant data at the cost of data accuracy, where some data generated by the sensing process might be an outlier. Thus, it is important to conserve the aggregated data's accuracy by performing an outlier data detection process before data aggregation is implemented. This paper concerns evaluating multivariate outlier detection (MOD) analysis on aggregated accuracy of data generated by a forest fire environment using OMNeT++ and performing the analysis in MATLAB R2018b. The findings of the study showed that the MOD algorithm conserved approximately 59.5% of aggregated data accuracy, compared with an equivalent algorithm, such as the FTDA algorithm, which conserved 54.25% of aggregated data accuracy for the same event.

FDPA internet of things system for forest fire detection, prediction and behaviour analysis

“…The major goal of establishing a WSN is to communicate data among the relevant entities, and star topology and mesh topology-based networks have been proposed in many articles due to the differences in their systems' features. A sensor node was recommended at three different heights to undertake different parameter measurements to identify fire conditions; however, it has been proposed by a few experts to prevent the rate of false alarms by shielding sensor nodes from direct sunlight [5,12,13].…”

A Literature Review on Outlier Detection in Wireless Sensor Networks

“…Statistical-based approaches like Copula-Based Probabilistic Multivariate (CBPM) [20] and Multivariate Outlier Detection (MOD) [38] have also been proposed. The CBPM technique considers the dependence between measurements surpassing existing statistical methods and is implemented in three stages: estimation, distributed detection, and outlier classification.…”

“…This latter, in most cases, is generated entirely based on certain statistical distributions, combining the injection of artificial anomalies, manual labeling, and even normalization processes. Cases such as [29] generate synthetic datasets based on real datasets, whereas in [38] only the former type is used. In Ref.…”

“…Table S3 shows that regarding the effectiveness of the proposed techniques, detection capabilities in most cases range between 85%-100%. It can also be observed [41,38] experiments, experimentation was not precisely aimed at determining the effectiveness of outlier detection but at measuring other aspects related to the proposed technique. In studies [40] and [50] no quantitative results of any kind are reported.…”

“…In the case of OFN [42], the selection of suitable parameters and sensitivity to variations in the density function can also affect its performance and effectiveness. The MOD approach [38] not only has a centralized architecture that can affect scalability and efficiency, but also requires the selection of appropriate parameters. The ANN-based technique [41] could face problems of computational complexity and energy consumption due to the nature of neural networks.…”