Both Artificial Intelligence (AI) and Machine Learning (ML) techniques have undergone a spectacular growth in the last decades, promising to help in the problem solution of various research areas. This article focuses on the applicability of Machine Learning techniques in the identification and prevention of network overload. To do this, we have collected real traffic matrices from the Abilene American network, these have been labeled as risky or not, and then fed into a set of Supervised Learning algorithms to see if they have the ability to forecast network overload situations. Indeed we observe that Quinlan’s C5.0 classification tree outperforms at detecting risky situations in which possible network overload can occur upon link failure (what-if analysis), showing accuracy values above 99% and Cohen’s kappa coefficient above 94%. This way, network operators can anticipate problematic network situations and prepare preventive countermeasures to mitigate their effects before they actually happen.