Android malware analysis in a nutshell.

Area of Science:

• Computer Science
• Cybersecurity
• Machine Learning

Background:

• Existing Android malware analysis often overlooks critical factors influencing vision-based detection.
• A comprehensive understanding of these factors is needed to enhance predictive system performance.

Purpose of the Study:

• To propose a comprehensive analysis model for Android malware focusing on vision-based factors.
• To empirically investigate the impact of various factors on the security and complexity of malware analysis.

Main Methods:

• Utilized 22 Convolutional Neural Network (CNN) algorithms, including 21 established and 1 novel algorithm.
• Employed diverse file types converted to images and two benchmark Android malware datasets.
• Conducted a deep empirical study evaluating models from security and complexity perspectives.

Main Results:

• Identified specific factors that significantly influence Android malware analysis performance.
• Demonstrated substantial improvements in accuracy (131.29%), F1-score (236.44%), precision (192%), and recall (131.29%) by adjusting key factors.
• Observed significant impacts on complexity metrics like testing time, CPU usage, storage size, and pre-processing speed.

Conclusions:

• The proposed model provides crucial guidance for researchers and developers in building efficient malware analysis systems.
• Highlights the critical importance of considering identified factors for optimizing both security and operational efficiency in Android malware detection.