Random Forest Prediction of WLAN Throughput using Communication Logs and Channel Occupancy Rate

With the advent of high-speed and low-latency wireless communications such as 5 G and $\mathrm{Wi}-\mathrm{Fi} 6$, more and more users are using them for large traffic applications such as high-resolution video streaming and cross reality (XR) services. However, the quality of wireless communications can be degraded by interference and other factors, which can reduce the quality of such services. Prediction of wireless communication quality is a useful method to deal with this problem. A method has been proposed to avoid degradation of quality of service (QoS) by predicting throughput using time-series data of throughput, which is one of the indicators of wireless communication quality. However, existing methods require constant measurement of throughput, which increases the traffic load and may degrade the quality of service. In this paper, we propose a method for predicting wireless communication quality without traffic load for measuring throughput by using communication logs. The proposed method uses received signal strength indicator (RSSI), channel occupancy rate (COR) and modulation and coding scheme (MCS) as inputs. The learning is performed using a random forest to predict the throughput that can be potentially transmitted from a node with new connection. Through experimental evaluation, we have confirmed that we can predict the potential throughput with an average error of about 16% in areas where RSSI is greater than -60 dBm by using communication logs and COR.