Our analysis leveraged machine learning algorithms (in particular, we evaluated two popular off-the-shelf classifiers: KNN and Random Forest) to classify the programs of our dataset as malware or goodware and to identify the potential concept drift that occurs when the subject of a classification scheme evolves as time goes by. This dataset is representative of the Brazilian cyberspace and contains profiles of known-bad and known-good programs based on binaries’ static features. We also share a labeled dataset of this Brazilian malware set to allow other experiments and comparisons by the community.
In this work, we present an analysis of thousands of malware samples collected in Brazilian cyberspace along several years, including their evolution and the impact of this evolution on malware classification. Analysis of these properties can lead to improved management of the digital commons. Ruling out technological constraints and workweek conventions, the study suggests that these covarying patterns come about because UCE senders strategi- cally exploit the unique features of the on-line commons, including instantaneous feed- back, information transparency, identity misrepresentation, and technological progress. The distribution of UCE messages reveals a cyclical trend, peaking in mid-week and subsiding on weekends, that can be explained in part by the trend of regular e-mail messages-an unanticipated finding given the difference bet ween UCE and ordinary e- mail communication. The absence of a growth trend in UCE message volume raises questions about the sampling methodologies underlying media reports about spam. This study, based on an analysis of 47 million inbound e-mail messages drawn from a cross-section of e-mail inbox owners over a one-year period, characterizes resource overuse in the e-mail commons.
Unsolicited commercial e-mail (UCE) is a significant problem of the digital commons, but there has been little empirical analysis of proposed solutions and underly- ing mechanisms.
0 kommentar(er)
