Detecting the online romance scam: Recognising images used in fraudulent dating profiles

The online romance scam is a scam with a high financial impact as well as a high emotional impact on the victims. Neither law nor awareness campaigns have proven to be effective against this scam, so a technical solution might be needed. This paper looks at a technique which can be used in such a technical solution. We present a classifier which is trained to recognise images used in the romance scam by using the occurrence of these images on the web. Besides this a dataset is constructed consisting out of images used in the romance scam complemented with normal images of people. We achieve an accuracy of 92.4% combined with a false negative rate of 19.7% using a random forest classifier. Although the results are promising, further research is needed to amongst others lower the false negative rate before this technique can be implemented in an end-user tool.