Estimating End-User Throughput Using Service Provider Cell Traces Via Gradient Boosting

The adoption of 5G networks has enabled supporting applications that require high bandwidths and low latencies. Service providers need to manage their resources efficiently to avoid service interruptions. Therefore, being aware of customer's experienced bandwidth is of paramount importance. Utilizing the traces collected on the service provider's side to estimate the experienced bandwidth on the user's side is a problem that was not studied in the literature. Moreover, the traces collected by the service provider are sparse and missing a large number of values to be reliable in predicting the user's experienced bandwidth. In this thesis, we focus on accurately imputing missing values in the collected traces, and consequently, we build a Regularized Gradient Boosting model to predict the user's throughput using traces that are exclusively collected from the service provider's resources. Our approach shows that using our imputing and prediction approaches, we can accurately estimate the user equipment's throughput.