Analysis of attacks on cloud computing with deep learning method

Cloud computing is a service model where computing resources such as servers, storage space, network infrastructure, data centers, and software services are provided over a shared network. This service model offers users flexibility, scalability, and cost savings. Despite the potential benefits of cloud computing, concerns about security have prevented individuals and businesses from fully embracing it. As a result, cloud computing security has become a new research topic. Deep learning holds significant potential in the field of cloud computing security. As the complexity and volume of security threats increase, deep learning algorithms can play an effective role in detecting security vulnerabilities, preventing attacks, and ensuring data privacy within cloud systems. Deep learning-based analysis methods can examine large datasets to identify abnormal behaviors and enable prompt responses. These analyses often utilize neural network architectures. In this thesis study, comprehensive research was conducted on the security issues in cloud computing. The challenges faced in the infrastructure-as-a-service model of cloud computing also affect other service models. Therefore, using deep learning techniques, various attack types and normal network traffic were monitored using extensive datasets named UNSW-NB15 and AWID, which represent network behaviors. Within the scope of the study, two models were created from neural network architectures commonly used in deep learning, namely Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM), which is a type of Recurrent Neural Network (RNN). In the first model, a hierarchical structure was employed where LSTM was used to learn the features of input data previously learned by CNN. In the second model, CNN and LSTM worked independently on the input data, and their outputs were merged in parallel. After training and testing these two models with the aforementioned datasets, the results were compared. It was observed that the hierarchical CNN-LSTM model had an accuracy rate of 82.24% on the UNSW-NB15 dataset, while the parallel CNN-LSTM model had an accuracy rate of 82.89%. It was observed that the hierarchical CNN-LSTM model had an accuracy rate of 83.29% on the AWID dataset, while the parallel CNN-LSTM model had an accuracy rate of 98.30%. Combining CNN and LSTM neural networks appears to be a powerful model building method. Both approaches have their own strengths and weaknesses. Therefore, it follows that the best choice depends on the specific problem.