Anti-virus technologies try to discover malicious web resources by downloading the resource content and evaluating its level of threat. Since there is a huge number of resources to be checked, it is infeasible to download and analyze all of them. The idea of this thesis is to efficiently decide whether an unknown content should be downloaded and checked just by analyzing its Uniform Resource Locator (URL) address. The URL is a way of addressing web resources by a text string, which can be further split into a host, path and query parts. In this thesis, the student will train some machine-learning classification technologies to estimate the input-URL probability of being malicious. The suitability of the technologies will be experimentally evaluated on a real-life dataset. The text part will describe the problem of classifying URL addresses, the state-of-the-art techniques, and the used technologies along with experimental evaluation.