A cheap and accurate delay-based IP Geolocation method using Machine Learning and Looking Glass

• Published in: 2023 IFIP Networking Conference (IFIP Networking) pp. 1 - 9
• Main Authors: Hong, Allen, Li, Yahui, Zhang, Han, Wang, Ming, An, Changqing, Wang, Jilong
• Format: Conference Proceeding
• Language: English
• Published: IFIP 12.06.2023
• Subjects: Costs; Delays; Geology; Glass; IP Geolocation; IP networks; Machine learning; Machine learning algorithms; Network Measurement
• ISSN: 1861-2288
• DOI: 10.23919/IFIPNetworking57963.2023.10186436

Predicting the geographical location of an IP host is a fundamental and valuable but long-standing challenge in the field of network research. Although delay-based methods have relatively high coverage and low time consumption, currently this type of method is not accurate enough and requires a large number of vantage points, making its cost high. In this paper, we propose a novel delay-based framework to make IP geolocation more accurate and cheap. Firstly, we collect 373 Looking Glass with known geographical addresses and overcome the high cost problem by using them as vantage points. Secondly, we make the prediction of geographical coordinates more accurate by using the machine learning algorithm and regional information of the target IP. Finally, we propose a method based on machine learning to supplement missing values in the delay data and improve the accuracy of geolocation successfully. Our experiment results validate the feasibility and improvement of our method. Using our method, we have an average error of 69.49 km for the geolocation of our test set, which is approximately 160 km less than the state-of-art work.