Ensemble Recurrent Graph Neural Networks for Availability Prediction in Cellular Networks

The self-organizing network mechanism based on machine learning technologies enhanced cell capabilities in the 5G RAN intelligent controller. In this paper, the cell availability prediction based on ensemble recurrent graph neural networks in cellular networks was proposed to optimize the QoS of mobile networks and improve the customer experience. Additionally, 5G RAN intelligent controller supported the self-healing network mechanism and followed the standard specifications of the open RAN architecture, 3GPP LTE key performance indicator, and Broadband Forum data model. In addition, the quality metrics, KPIs, and device logs were introduced into the learning model input factor for cell availability prediction. Experimental results show that the ensemble graph neural networks effectively predict cell availability better than traditional multilayer perceptron structure, and the learning model accuracy was more than 95%.