Introducing and Comparing Recent Clustering Methods for Massive Data Management in the Internet of Things

Abstract

The use of wireless sensor networks, which are the key ingredient in the growing Internet of Things (IoT), has surged over the past few years with a widening range of applications in the industry, healthcare, agriculture, with a special attention to monitoring and tracking, often tied with security issues. In some applications, sensors can be deployed in remote, large unpopulated areas, whereas in others, they serve to monitor confined busy spaces. In either case, clustering the sensor network’s nodes into several clusters is of fundamental benefit for obvious scalability reasons, and also for helping to devise maintenance or usage schedules that might greatly improve the network’s lifetime. In the present paper, we survey and compare popular and advanced clustering schemes and provide a detailed analysis of their performance as a function of scale, type of collected data or their heterogeneity, and noise level. The testing is performed on real sensor data provided by the UCI Machine Learning Repository, using various external validation metrics.