Advanced Malware Detection: Integrating Convolutional Neural Networks with LSTM RNNs for Enhanced Security
Malware, or malicious software, is a serious threat to people, businesses, and the cybersecurity environment as a whole. Its purpose is to disrupt, damage, or obtain unauthorized access to computer systems. The ability to accurately classify and identify different types of malware is very important...
Saved in:
| Main Author: | |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
College of Computer and Information Technology – University of Wasit, Iraq
2024-12-01
|
| Series: | Wasit Journal of Computer and Mathematics Science |
| Subjects: | |
| Online Access: | http://wjcm.uowasit.edu.iq/index.php/wjcm/article/view/288 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Malware, or malicious software, is a serious threat to people, businesses, and the cybersecurity environment as a whole. Its purpose is to disrupt, damage, or obtain unauthorized access to computer systems. The ability to accurately classify and identify different types of malware is very important in developing effective defense mechanisms and reducing possible risks In order to classify malware from photos, this paper presents a novel approach that combines the capabilities of an LSTM architecture with the convolutional neural network AlexNet. We began with preprocessing the data, which included resizing the images for compatibility with the network architecture. Then, we used AlexNet to extract powerful and meaningful features from the malware images. Although we extracted 1,000 features, we trimmed the list to 120 features using linear discriminant analysis for more efficient and effective classification. Finally, we trained an LSTM network with the extracted features. The images used in our experiments contained malware from nine different families. To evaluate the performance of our proposed approach, we conducted experiments on the MaliMG dataset, which includes a diverse range of malware samples. The obtained results show the effectiveness of the proposed method. The training accuracy reached a significant value of 99.80%, which shows the ability of our model to accurately learn patterns and features of malware images. Moreover, the evaluation of the test dataset yielded a remarkable accuracy of 99.49%, which highlights the robustness and generalizability of our approach.
|
|---|---|
| ISSN: | 2788-5879 2788-5887 |