Multiclassificador para a detecção de Outliers em dados gerados por sensores de monitoramento ambiental

Abstract

In the context of the Internet of Things (IoT), one of the applications are smart cities, in which technology is employed in order to improve the quality of life of its citizens. In this scenario, there is an intense data generation from sensors, which provide important information about their surroundings and can be used in decision making. However, new challenges also emerge, such as the presence of outliers, that is, values that differ drastically from others in the same dataset, which can lead to misinterpretations. In view of this, the current work presents the construction of a multiclassifier algorithm for outlier detection in data obtained from three envi ronmental monitoring IoT sensors. The multiclassifier is composed of two statistical techniques (Zscore and Modified Zscore) and one of clustering (K-Means). The methods are evaluated and compared using performance metrics of sensitivity, precision, specificity and accuracy. It is found that Zscore and Modified Zscore exhibit better performance in identifying abnormali ties in unimodal distribution data, while K-Means has higher efficiency in bimodal distribution data. Therefore, by gathering the responses in the multiclassifier, they complement each other, yielding a more robust outlier detection system with better performance metrics