Improving IP geolocation databases based on multi-method classification

Querying IP geolocation databases directly with a host's IP address is a common and convenient way to determine the geographic location of that host. However, there are usually a multitude of unsatisfactory "null" replies and data inconsistencies in the query results. Thus, this paper applies a novel approach to calibrate geolocation databases. We use a variety of classifiers and employ time delays as features to automatically reveal the relation between distances and time delays. To begin with, target Internet hosts with authentic locations are obtained by a strategy of taking advantage of current databases. Then the time delays measured from a group of fixed servers to target Internet hosts are collected comprehensively from an online website and employed as features. Finally, two multi-method-based models are proposed to find out the locations of Internet hosts. The experiment results show that our method achieves 99% precision at the province level as well as 81.65% precision at the city level on the self-collected data set. The results highlight that time delays can be well used as features and the model we proposed is a powerful tool to improve the quality of IP geolocation databases.