Dynamic Feature-Based Detection of Malware in Non-Executable Files Using a 1D Convolutional Neural Network

Abstract

With the exponential growth of data, the threat posed by malware has been on the rise. Particularly, nonexecutable documents like PDF files can be highly dangerous due to the difficulty in detecting them and the lack of awareness among users regarding such malicious attacks. In this research article, we present a one-dimensional convolutional neural network (1D-CNN) specifically designed for detecting malware in PDF files. To accomplish this, we gather a dataset comprising both malicious and benign PDF files (Evasive-PDFMal2022). We extensively analyze the structure of the input data and explain how our proposed network is tailored to leverage the inherent characteristics of the data. Our network is designed to capture meaningful patterns among spatial clues obtained from the data, enabling it to predict whether a given byte sequence contains malicious actions. Through experimental evaluations, we demonstrate that our proposed network surpasses several state-of-the-art machine learning models and other networks with different configurations, affirming its superior performance in malware detection.