Abstract

This paper presents, describes and evaluates the Machine Learning Performance Monitor (MLPM), an innovative Machine Learning (ML) approach to forecast and extrapolate the performance of several network features (e.g., latency, throughput) in a Multipath TCP (MPTCP) subflow pool. MLPM uses linear regression to predict the performance of network features along with Artificial Neural Network linear classifier to choose the best subflow (i.e., network path) capable of delivering the best performance to a given set of the network features. Results show that MLPM delivers better performance in terms of throughput and latency compared to existing schemes as it improves the MPTCP scheduler performance.