The emerging development of remote sensing and control technology gives rise to new types of real-time networking applications that require "pro-active" quality of service guarantees. In particular, it is important to predict the probability that the available end-to-end bandwidth drops below a specified quality-critical threshold for a given period of time. We refer to such a probability as to a "crossing probability".

In this work we develop a non-parametric approach for prediction of the "crossing probability" for an arbitrarily taken Internet path. We also provide a "self-check" and therefore are able to evaluate the accuracy of our prediction. Experimental verification of prediction accuracy is based on real-life measurements collected from multiple Internet paths.