Near Field Communication (NFC) has enabled mobile phones to emulate contactless smartcards, which also makes them susceptible to relay attacks. One of the methods proposed to counter such attacks is proximity detection via ambient sensors. The idea is that a mobile phone and transaction terminal are in a similar environment (as captured by the sensors) during a genuine transaction, whereas the environments are likely to be different during a relay attack, i.e. a malicious transaction.
Over the course of my internship, I used Deep Learning (DL) methods to perform proximity detection on a dataset consisting of ambient sensor readings taken from mobile phones and transaction terminals during contactless transactions. The dataset resulted from a previous experiment and has already been analysed with threshold-based similarity metrics and Machine Learning techniques. While the evaluated DL models performed slightly higher overall, the results are not yet accurate enough to use in a commercial setting.
