Malware Analysis Using Low-Level Behavioral Artifacts

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing malware classification and clustering systems rely on high-level attributes and limited sample sets, leading to reduced precision and inconsistencies, and often use a single algorithm, which restricts their effectiveness.

Innovation Solution

The system employs behavior-based automated malware analysis and classification using a large set of low-level operational artifacts and an extensible set of algorithms, allowing for flexible algorithm selection and tuning to improve accuracy, with features extracted from file system, registry, and network artifacts, and utilizing algorithms like SVM and decision trees for classification, and hierarchical clustering for grouping.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If high-level attributes are used for malware classification, then the system is simpler to implement, but the precision and granularity of classification deteriorates

Engineering Contradiction:
Improvesystem complexityVSAvoidclassification precision
Core Design Contradiction:
Device complexityVSMeasurement precision

Solution Approach 1:

The patent segments malware analysis into multiple levels: low-level artifact capture (file system, registry, network artifacts), feature extraction from these artifacts, and then classification/clustering based on extracted features. This segmentation allows the system to maintain simplicity in each component while achieving high overall precision through the combination of detailed low-level observations with structured analysis pipelines.

Inventive Principle:
Principle #1Segmentation

2Device complexity

If a single algorithm is used for malware clustering, then the system is easier to implement, but the effectiveness and adaptability of results deteriorates

Engineering Contradiction:
Improvealgorithm complexityVSAvoidalgorithm adaptability
Core Design Contradiction:
Device complexityVSAdaptability or versatility

Solution Approach 1:

The patent implements a multi-algorithm framework where multiple clustering algorithms (hierarchical clustering, k-means, DBSCAN) and classification algorithms (SVM, decision trees, random forests) are integrated into a single system. The system can select and apply different algorithms based on the specific malware dataset and analysis requirements, making the system universally applicable to various malware families and scenarios while maintaining ease of implementation through a standardized interface.

Inventive Principle:
Principle #6Universality (Multi-functionality)

3Productivity

If a small sample set is used for attribute capture, then the analysis is faster, but the consistency and reliability of results deteriorates

Engineering Contradiction:
Improveanalysis speedVSAvoidresult consistency
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The patent performs preliminary action by capturing comprehensive low-level artifacts (file system operations, registry modifications, network communications) during the malware execution phase before classification occurs. This preliminary capture of detailed operational data ensures that sufficient information is available for reliable analysis, while the subsequent feature extraction and algorithmic processing maintain analysis speed by working with condensed feature representations rather than raw artifact volumes.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP2916256B1Systems and methods for behavior-based automated malware analysis and classification
Publication Date: 2019.11.20 VERISIGN INC
  • EP2916256B1 patent drawingFigure 1
  • EP2916256B1 patent drawingFigure 2
  • EP2916256B1 patent drawingFigure 3

AI summary

Embodiments relate to systems and methods for behavior-based automated malware analysis and classification. Aspects relate to platforms and techniques which access a set of samples of malware (102), and extract or capture a set of low-level behavioral artifacts (150) produced by those samples. The low-level artifacts can be used to organize or identify a set of features, based upon which the sample can be classified and/or clustered into different labels, groups, or categories. The artifacts and/or features can be analyzed by one or more selectable algorithms, whose accuracy, efficiency, and other characteristics can be compared to one another for purposes of performing a classification or clustering task. The algorithm(s) can be selected by a user to achieve desired run times, accuracy levels, and/or other effects.