WSCISOM: wireless sensor data cluster identification through a hybrid SOM/MLP/RBF architecture

Networks of wireless sensors are very popular devices for monitoring and collecting information about phenomena in many aspects of life. While very versatile and widely applicable, there are few key issues related to the operation of wireless sensors as well as the processing of information collected by them. In this paper, we focus on wireless sensor network (WSN) organization and protocols, energy consumption as related to information exchange and calculations, and making sense and applying the concluded decisions by the WSN. In addition to the clustering technique—we are utilizing modified self-organizing map (SOM)—we propose a hybrid multilayer perceptron (MLP) and radial basis functions (RBF) neural network to analyze and classify the possible routes taken by devices activating our WSN. The results demonstrate that the SOM modifications made with energy savings in mind perform very well and provide a quality input for the MLP/RBF classifier. The final goal of determining all possible areas of activity within an input space of interest is successfully achieved as demonstrated by the experiments.