Mitigating Measurement Failures in Throughput Performance Forecasting

Network monitoring services are performed by several companies and Internet Providers (ISP), which provide results of regular performance tests, such as throughput, loss, and delay, among others. These measurements help to understand the network’s behavior and obtain information for strategic planning. However, when carrying out the measurements planned during network monitoring, failures may occur, which makes it difficult to carry out more complex activities, such as forecasting network performance. Within this context, this article presents a resilient and adaptive model for forecasting network performance, which includes the identification of measurement failures and applying data imputation techniques to adapt the data for the forecasting process (based on Neural Networks and Time Series Analysis). The experiments, using real data from the National Education and Research Network (RNP), show that the proposal can achieve high accuracy in forecasting with imputed data and outperform other existing forecasting approaches.