Abstract
The multiplied development of cyber threats and types of malwares has triggered augmented usage of the artificial intellect (AI) and machine learning (ML) in intelligent malware detectors. The study gives a survey and a performance evaluation of ML and AI-based malware detection methods, reports on the research on the statical, dynamic, and hybrid models of malware detection on various operating systems, including Windows, Android, IoT, and cloud systems. Some of the algorithms that are assessed in the study include Random Forest, Support Vector Machine (SVM), Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), Graph Neural Networks (GNN), and Transformer-based architectures. Experimental comparison provides also the best accuracy of 98.7% and 98.1% of Transformer and GNN-based models, respectively, which is superior to the traditional ML models, including SVM (89.5%) and Random Forest (92.3%). Hybrid CNN-LSTM models were also found to be very efficient in detection with a 97.2 percent accuracy score. The findings show that deep and hybrid models are better in terms of adaptability and precision, but with increased computational expenses. Moreover, the implementation of explainable AI (XAI) increased the model resiliency up to 16.2 better interpretability and security. In general, this study demonstrates that AI has the potential to deeply change the nature of cybersecurity, and it is necessary to design the explainable, adversarial, and scalable malware detectors as the future of digital protection.