FILM: Filtering and Machine Learning for Malware Detection in Edge Computing.

Area of Science:

• Computer Science
• Cybersecurity
• Machine Learning

Background:

• Malware detection models using static-analysis features struggle with misclassification on edge devices.
• Resource-constrained environments like IoT devices require efficient and accurate malware detection.

Purpose of the Study:

• To develop a novel detection method for edge computing that improves upon existing binary classification models.
• To enhance the accuracy, precision, and recall of deep learning models for malware detection.

Main Methods:

• A new detection method is proposed that classifies files into benign, malicious, or unpredictable categories.
• Existing deep learning models are adapted by appending a sigmoid function to assess prediction confidence.
• The method filters predictions based on model confidence to reduce errors from ambiguous features.

Main Results:

• Experiments on real malware datasets show significant enhancements in accuracy, precision, and recall.
• Accuracy improved from 0.96 to 0.99 for the enhanced deep learning models.
• The method successfully identifies some files as unpredictable, allowing for further cloud-based analysis.

Conclusions:

• The proposed method effectively enhances existing machine learning models for malware detection in edge computing.
• Classifying files as unpredictable provides a mechanism for handling ambiguous cases, improving overall system reliability.
• This approach offers a more robust solution for cybersecurity in resource-constrained edge and IoT environments.