Predicting data traffic in cellular data networks

Sammanfattning: The exponential increase in cellular data usage in recent time is evident, which introduces challenges and opportunities for the telecom industry. From a Radio Resource Management perspective, it is therefore most valuable to be able to predict future events such as user load. The objective of this thesis is thus to investigate whether one can predict such future events based on information available in a base station. This is done by clustering data obtained from a simulated 4G network using Gaussian Mixture Models. Based on this, an evaluation based on the cluster signatures is performed, where heavy-load users seem to be identified. Furthermore, other evaluations on other temporal aspects tied to the clusters and cluster transitions is performed. Secondly, supervised classification using Random Forest is performed, in order to investigate whether prediction of these cluster labels is possible. High accuracies for most of these classifications are obtained, suggesting that prediction based on these methods can be made.