Android malware detection method based on highly distinguishable static features and DenseNet.

Area of Science:

• Computer Science
• Cybersecurity
• Machine Learning

Background:

• The proliferation of malware poses a significant threat to the Android mobile ecosystem.
• Existing malware detection methods often suffer from low accuracy due to reliance on single features or inefficient use of multiple features.
• Large feature sets in multi-feature methods lead to high computational overhead, while inadequate feature selection compromises detection performance.

Purpose of the Study:

• To develop an accurate and efficient method for Android malware detection.
• To address the limitations of single-feature and multi-feature detection approaches.
• To reduce computational overhead while enhancing detection accuracy.

Main Methods:

• Proposed an Android malware identification method utilizing seven types of static features.
• Implemented a three-level feature selection process to identify highly distinguishable features.
• Employed a fully densely connected convolutional network (DenseNet) for efficient feature leveraging in detection.

Main Results:

• Reduced the feature set size by approximately 97% with a minimal accuracy drop of 0.45%.
• Achieved over 99% accuracy across various machine learning models, with the DenseNet model reaching 99.72% accuracy.
• Demonstrated superior performance compared to general machine learning models, state-of-the-art neural networks, and other multi-feature detection methods, with reduced training costs.

Conclusions:

• The proposed method effectively identifies Android malware with high accuracy and efficiency.
• The multi-level feature selection and DenseNet-based approach significantly optimize the detection process.
• This technique offers a robust solution to combat the increasing volume of Android malware.