Malicious File Detection Method Using Machine Learning and Interworking with MITRE ATT&CK Framework
GwangHyun Ahn, Kookjin Kim, Won-Hyung Park, Dongkyoo Shin
Abstract
With advances in cyber threats and increased intelligence, incidents continue to occur related to new ways of using new technologies. In addition, as intelligent and advanced cyberattack technologies gradually increase, the limit of inefficient malicious code detection and analysis has been reached, and inaccurate detection rates for unknown malicious codes are increasing. Thus, this study used a machine learning algorithm to achieve a malicious file detection accuracy of more than 99%, along with a method for visualizing data for the detection of malicious files using the dynamic-analysis-based MITRE ATT&CK framework. The PE malware dataset was classified into Random Forest, Adaboost, and Gradient Boosting models. These models achieved accuracies of 99.3%, 98.4%, and 98.8%, respectively, and malicious file analysis results were derived through visualization by applying the MITRE ATT&CK matrix.