Augmenting Network Performance Datasets with Weather, Sports, and Social Media Data for Improved Predictions

Understanding network performance enables network providers to manage their network better. Network performance degradation can lead to network service issues causing monetary loss and customer churn for the network providers. Accurate network performance prediction potentially enables proactive resource allocation to attenuate the anticipated network performance degradation and associated service issues. Previous literature attempted to predict network performance using historical network data. However, real-world network performance is impacted by various external factors. Existing literature fails to consider such external factors that can improve the understanding and predictions of the network performance. This thesis aims to examine if external factors can improve the network performance understanding and predictions. To this end, we inspect the correlation of network performance data with various external data sources such as weather parameters, sports events, and social media posts. Then, we perform network performance data augmentation using the contextual information in such external data. We investigate the network performance prediction improvements using Recurrent Neural Network (RNN) with Long Short Term Memory (LSTM) units after data augmentation. Predictive experiments with data augmentation using NFL sports events highlight a 23% improvement in the network performance predictions. Data augmentation using other external sources considered fails to improve the network performance predictions