Peak Hour Performance Prediction based on Machine Learning for LTE Mobile Cellular Network

Performance prediction is one of the main challenges in assessing the quality of any mobile cellular network. The prediction allows the operator to be aware of future network behaviors and thus take corrective actions to improve the performance network, and consequently the users Quality of Experience. However, this prediction can be affected by traffic higher demands, such as in peak hours of the day, specific days of the week and holidays. The present work uses three Machine Learning techniques for the prediction of LTE network performance. A database was built with information collected at peak hours of the day during two months. The CRISP-DM methodology was used as a model for data mining, and finally the Machine Learning models were evaluated using statistical metrics. This allowed to establish the Gaussian Regression Process as the model that best fits the proposed scenario, compared with two benchmark models, Support Vector Machines and Robust Linear Regression.