Enhancing cloud security using attention-based improved RegNet with hybrid vigenere-shift transposition algorithm

Cloud computing services have become essential daily, gaining popularity across several sectors due to their intrinsic qualities, including scalability and adaptability. However, despite its advantages, cloud networks are susceptible to security threats, including insider threats, data breaches, and illegal access due to their shared infrastructure, necessitating robust security mechanisms. One of the most effective defences is an Intrusion Detection System (IDS), which monitors network traffic and system activity to identify harmful threats. However, traditional IDS strugglesto accurately identify attacks due to the massive volume of cloud traffic and the complexity of modern intrusions. To tackle these challenges, this paper proposes a novel Deep Learning (DL)-based ID called Attention-Based Improved Regularization Network (ABIRegNet) with an encryption technique to enhance cloud data security. The intrusion dataset undergoes preprocessing to enhance quality by removing noise and normalizing. To mitigate the class imbalance problem, Adaptive Synthetic (ADASYN) technique is used. Next, we use the Variational Autoencoder (VAE) to select the relevant features, reducing computational complexity. The chosen features are sent into the proposed ABIRegNet to detect whether the data is normal or malicious. The ABIRegNet model integrates a Multi-Pooling Channel Attention (MPCA) mechanismto enhance intrusion detection by focusingon significant features. Furthermore, Parrot Optimization Algorithm (POA) is used to optimize the model hyperparameters for improved performance. After classification, normal data is encrypted utilizing the Hybrid Vigenere-Shift Transposition Algorithm (HVSTA), which combines the Vigenere Cipher (VC) with the Cyclic Shift Transposition Algorithm (CSTA) to securely store data on a cloud server. The suggested approach is implemented in Python. The proposed method has been analyzed on two datasets, achieving accuracies of 99.37% on CICIDS2017 and 99% on NSL-KDD. Experimental results illustrate that the suggested methodoutperforms state-of-the-art methods, indicating strong potential for enhancing intrusion detection and cloud data security.