False Positive Mitigation in Behavioral Malware Detection Using Deep Learning

The malicious software is in a continuous development and the anti-malware technologies are advancing as well to keep up. There are proactive detection technologies, based on the analysis of a sample behavior, that succeed in detecting zero-day malware, the downside being the false positives rate. The current paper proposes an approach for mitigating the false positives by introducing a deep learning classifier. This classifier provides a "second opinion" for the samples that would have been detected by the current state of the art approach. The proposed approach is able to reduce the false positives rate by 97%, while only losing 12% of the legitimate detection.