Cluster-Based Time Series Modelling and Forecasting of Mobile Network Traffic

In the ever-evolving digital age, advancements in communication and network technologies have revolutionized the digital way of interaction, which is driven by the rapid evolution of cellular networks. With the continuous improvements in the mobile network generations, a vast number of devices have emerged in recent years. In the meantime, the high volume of cellular traffic demands enhanced, reliable, and streamlined services. Addressing these growing demands requires an effective way of a proactive approach to allocating cellular network resources. One of the most important components of the network resource allocation management system is the forecast of network traffic, which is highly complicated in terms of accuracy and reliability. This study presents a cluster-based approach, known as within-cluster sum of squares (WCSS), using K-means clustering algorithms to aggregate the data points from the real-world Call Detail Records(CDR) dataset that uncover similar patterns of internet traffic. This approach allows for the creation of a more organized and consistent analysis process, leading to more accurate forecast of network traffic. The selected time series models of ARIMA, SARIMA, and LSTM were employed in the clustered dataset and evaluated against error metrics of MSE, RMSE, and MAE. A comparative analysis of internet traffic forecasting models showed that ARIMA outperformed both SARIMA and LSTM in terms of accuracy. The findings of the study mark a promising step, by introducing the cluster-based approach for time series analysis of mobile Internet traffic.