Sparse attention with residual pyramidal depthwise separable convolutional based malware detection with optimization...

Area of Science:

• Cybersecurity
• Machine Learning
• Cloud Computing

Background:

• Malware poses significant risks to cloud system security and privacy.
• Traditional signature-based malware detection methods are inefficient against evolving threats.
• Existing API call analysis models face accuracy and classification challenges.

Purpose of the Study:

• To develop an advanced deep learning-based malware detection system for cloud environments.
• To address limitations of existing methods by utilizing image-based analysis of API calls.
• To improve the accuracy and efficiency of malware classification.

Main Methods:

• Converted API call data into 2D grayscale images for analysis.
• Employed image pre-processing using weighted mean and anisotropic filters.
• Utilized an integrated densely connected squeeze MobileNet v2 (Ef-DeSMob2) for feature extraction.
• Implemented sparse attention with residual pyramidal depth wise separable convolutional neural networks (SA:ResPyDSC) for classification.
• Fine-tuned classifier hyperparameters using a hybrid white shark beluga optimization algorithm (Hy-WBeOp).

Main Results:

• Achieved high accuracy (98.06%), precision (97.99%), recall (97.05%), and F1-score (96.08%).
• Demonstrated low error rates with MSE (0.08), RMSE (0.27), and MAE (0.21).
• Outperformed existing techniques in efficiency and accuracy for malware classification.

Conclusions:

• The proposed deep learning approach effectively classifies malware in cloud systems.
• Image-based analysis of API calls enhances detection reliability and reduces errors.
• This method establishes a robust system for protection against sophisticated cyber threats.