Genetic Machine Learning Approach for Link Quality Prediction in Mobile Wireless Sensor Networks

Abstract

Establishing adequate RF (Radio Frequency) connectivity is the basic requirement for the proper operation of any wireless network. In a mobile wireless network it is a challenge for applications and protocols to deal with connectivity problems, as links might get up and down frequently. In these scenarios, having knowledge of the node remaining connectivity time can avoid unnecessary or even unuseful control/data messages transmissions. The current paper presents the so-called Genetic Machine Learning Approach for Link Quality Prediction, or simply GMLA, which is a solution to forecast the remainder RF connectivity time in mobile environments. Differently from all related works, GMLA allows building connectivity knowledge to estimate the RF link duration without the need of a pre-runtime phase. This allows to apply GMLA at unknown environments and mobility patterns. Its structure combines a Classifier System with a Markov chain model of the RF link quality. As the Markov model parameters are discovered on-the-fly, there is no need of a previous history to feed the Markov model. Obtained simulation results show that GMLA is a very suitable solution, as it outperforms approaches that use geographical positioning systems (GPS) and also approaches that use link-quality prediction, such as BD and MTCP. GMLA is generic enough to be applied to any layer of the communication protocol stack, especially in the link and network layers.