Effective Android Malware Detection Using CNN and LSTM Model with GWO-Based Feature Selection

Abstract

Malware or malicious software represents one of the most significant cybersecurity threats, compromising the integrity, confidentiality, and availability of computer systems and networks. Traditional malware detection methods seek to locate specific malware samples and families to recognize harmful code, and can be located using traditional signature- and rule-based detection methods. The study aims at the creation of malware detectors based on deep learning and optimization algorithms. The present study will propose a more advanced malware detection system that employs deep learning models and the Grey Wolf Optimization (GWO) algorithm to select essential features. This model used data balancing Synthetic Minority Oversampling Technique (SMOTE), and the models were tested on the CICMalDroid2020 dataset. Using two different models were tested: Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) with Grey Wolf Optimization. The experimental findings indicate that the CNN-GWO model has an model achieves 93% accuracy, 92% precision, and 92% recall, and the LSTM-GWO model has an 90% accuracy, 90% precision, and 90% recall, respectively. These findings show that both models can be effective in detecting and classifying malware, but CNN has a few higher performance rates. The novelty of this method is the independent application of GWO-based feature selection to CNN and LSTM architectures, enhancing the detection accuracy and efficiency of these architectures in comparison to the latest research.