A comparative study of linear and non-linear dimensionality reduction for opcode-frequency malware classification

Key Points

• Malware classification accuracy varies with dimensionality reduction techniques used, showing notable differences between linear and non-linear methods.
• The highest accuracy reached 95.3% when non-linear techniques were applied in contrast to linear methods, indicating superior performance in modeling.
• This analysis utilized comparative methods to assess linear and non-linear dimensionality reduction capabilities on opcode-frequency data for malware classification.
• The findings suggest that non-linear dimensionality reduction may enhance malware detection systems, highlighting the importance of algorithm choice.