Improvising the Malware Detection Accuracy in Portable Document Format (PDFs) through Machine Learning Classifiers

Every time a spike is observed in cyber-attacks, a huge financial loss is incurred that has surpassed $2 trillion according to some estimates. Apart from monetary setbacks, the damages in terms of loss of credibility and quality of services are way more difficult to recover. PDF documents have been considered as innocuous source of data sharing with static meta-tags. However, the structure of PDF file contains different objects, features and characteristics that make it easier for the attackers to invade (by attaching the malicious code with tags). Current research is focused on analyzing the structure of PDF files, data-structures exploited and extracting feature tags with associated feature values of PDFs. A total of 17,884 documents were used, having two categories (8942 malicious documents and 8942 benign files).