Abstract
Although conveniences have accelerated due to the rapid growth of the internet and computer systems, ways to perpetrate cybercrimes have also been on the rise, particularly visible in the continuous malware attacks. This study takes a combined cloud computing and artificial intelligence methods to provide a hybrid deep learning-based architecture for the improvement of malware detection. For substantial classification from CNNs through feature extraction, the proposal contains transfer learning, adversarial training and ensemble methods. It is assured by standardizing the data from network traffic logs social data sets from EICAR and the public viral dataset archives. The experimental results on the Malimg dataset showed 97.8% accuracy which exceeded the accuracy of AlexNet and ResNet-50. The proposal shows a way of improving cybersecurity with some new and hybrid scalable and robust methods.