Particle Swarm Optimization of Machine Learning in Cognitive Radio e-Health Networks at Low SNR

The cognitive radio (CR) may play an important role in modern wireless communication systems such as ehealth systems. It can help remotely monitoring patients while they are moving in the hospital premises. CR can be used to relay the patients’ data to the healthcare data center without the need to purchase a dedicated spectrum by using unutilized spectrum bands within the hospital premises. The spectrum sensing (SS) is the core of the CR technology. Recently, machine learning (ML) algorithms have been applied to SS to achieve better spectrum detection performance. However, existing ML-based SS do not consider the issue of properly defining the ML algorithm parameters, which has a direct impact on the performance, particularly at low signal-to-noise ratio (SNR). In this paper, we propose to optimally find the parameters using particle swarm optimization (PSO). This results in the automatic configuration of the ML algorithm parameter rather than manually configuring them as the case with the related literature. We apply the proposed optimization on Extreme Learning Machine (ELM), K-Nearest Neighbors (KNN), Support Vector Machine (SVM), and Ensemble Classifier (EC) algorithms for SS. Our results show that PSO-optimized algorithms outperform the manually optimized algorithms by 3.55% to 7.15% in accuracy and 5.94 to 13.6% in the detection probability at false alarm probability equal to 0.1. The PSO-optimized EC algorithm is the most effective in satisfying the IEEE 802.22 requirements at low SNR.