Malware detection through low-level features and stacked denoising autoencoders

In recent years, the diffusion of malicious software through various channels has gained the request for intelligent techniques capable of timely detecting new malware spread. In this work, we focus on the application of Deep Learning methods for malware detection, by evaluating their effectiveness when malware are represented by high-level, and lowlevel features respectively. Experimental results show that, when using high-level features, deep neural networks do not significantly improve the overall detection accuracy. On the other hand, when low-level features, i.e., small pieces of information extracted through a light processing, are chosen, they allow to increase the capability of correctly classifying malware.