Trunked radio systems: traffic prediction based on user clusters

Studies of individual network users' behavior patterns may seem of little relevance to predicting the entire network's traffic. Clustering techniques, however, help bridge this apparent gap. In this paper, we analyze data collected from a deployed network and use clustering techniques to characterize patterns of individual users' behavior. A network traffic prediction approach is then developed based on user clusters. We analyze three months of continuous network log data from the E-Comm network, an operational trunked radio system. After extracting traffic data from the raw data logs, we identify user clusters by employing the AutoClass tool and the K-means algorithm. Based on the identified user clusters, we use the seasonal autoregressive integrated moving average (SARIMA) model to forecast the network traffic by aggregating the predicted traffic of each user cluster. The predicted network traffic shows good agreement with the collected traffic data.