DEEP LEARNING-BASED INTRUSION DETECTION WITH RESOURCE-AWARE AUTO-SCALING FOR SECURE CLOUD COMPUTING

Abstract

Cloud computing infrastructures have become a part of the modern digital ecosystem that supports, For enterprise usage such as high availability, elasticity, and operational efficiency. With this increasing dependency, cloud environments have simultaneously become major targets for advanced cyber-attacks like Denial of Service (DDoS), brute-force and more. Attacks (attempts) and botnet attacks. Traditional intrusion detection systems (IDS), are usually limited to static signatures or rule-based logic, and are often not able to detect attack patterns were emerging or dealing with the high dimensional nature of real network traffic. Moreover, current security solutions don’t typically measure the impact of malicious traffic on cloud resources. CPU utilization, especially in regard to CPU load, is critical for the quality of service. Providing automatic scaling mechanisms. This work, a unified deep learning based framework is proposed to simultaneously ad Design cyber threat detection and resource consumption forecasting in cloud systems. Using the CICIDS2017 dataset, Two types of networks were implemented, namely Long Short Term Memory(LSTM)andConvolutional Neural Networks (CNN) for binary traffic classi fication. The LSTMmodelwasableto obtain an accuracy of about 97% which is better than the CNN. An important reason why model is found at 96% is that it is capable of modeling the temporal relationships between sequential network flows. We modeled the operational effects of attacks with a Random Forest regression model. has been trained to predict CPU utilization by malicious traffic intensity with R² score of 0.90. This high predictive capabil ity clearly shows that there is a relationship between attack patterns. With the development of cloud security analytics, it will be more feasible to integrate security analytics resources, and security analytics will be more widely used in the future. In combination with predictive resource management. A simulated auto-scaling was then developed that used the predicted CPU utilization. A mechanism to mimic the cloud platform behaviour when it is subjected to load fluctuations during an attack. Scale out events were started successfully if CPU thresholds were exceeded, this is an illustration of the potential benefit of predictive, security-aware auto-scaling in minimizing performance impact degradation during cyberattacks.