Abstract: In order to predict the link quality accurately, a link quality prediction model was proposed to predict link quality for sensor networks based on improved least square support vector regression machine (LS-SVR). The rough set (RS) was introduced to reduce the link quality metrics so as to extract the effective characteristic metrics of the link quality. And the genetic algorithm (GA) was employed in LS-SVR to optimize the penalty factor and kernel width. Experiments show that compared with the experts advice-based prediction model, the proposed prediction model achieves better accuracy.

Key words: wireless sensor networks, link quality prediction, genetic algorithm, least squares support vector regression machine