ABSTRACT
The threat of malware looms large in today’s world. It is of paramount importance that malware detection strategies evolve at an appropriate speed. The recent boost in exchange of images on social media as well as popularity of IoT devices are significant contributors of the current malware landscape. The present-day malware detection products are based on intensive manual effort and thus consume a lot of time to detect malware. It is a burning need to be able to detect malware in less time. This will not just save time and thus negate malicious intent of malware developers; it will also save a lot of money involved in malware detection.
Training machines to detect malware can significantly reduce the time required to detect malware. These techniques can be reasonably reliable because by following a consistent process, all but the outliers can be detected. Once machines can be made to learn the process of detecting malware, the time required to detect malware will reduce drastically. Also, as the manpower required in the process of malware detection using machines will be a fraction of what is required now, the net cost of detecting malware will reduce drastically.
In this chapter, we develop and evaluate two machine learning classifiers, the former capable of detecting malicious JPEG files with 99.9% accuracy and the latter capable of detecting and classifying malicious ELF files into malware categories with 87% accuracy.
