Staked deep ensemble model for intruder behaviour detection and classification in cloud

• Published in: Multimedia tools and applications Vol. 83; no. 19; pp. 57861 - 57892
• Main Authors: Mohan, M., Tamizhazhagan, V., Balaji, S.
• Format: Journal Article
• Language: English
• Published: New York Springer US 14.12.2023; Springer Nature B.V
• Subjects: Algorithms; Belief networks; Classification; Computer Communication Networks; Computer Science; Data Structures and Information Theory; Feature extraction; Information systems; Intrusion; Intrusion detection systems; Multimedia Information Systems; Principal components analysis; Recurrent neural networks; Security; Special Purpose and Application-Based Systems; Track 6: Computer Vision for Multimedia Applications; Intruder; Deep Learning; Optimization; Cloud; Ensemble
• ISSN: 1573-7721; 1380-7501; 1573-7721
• DOI: 10.1007/s11042-023-17677-9

Computer systems, cloud networks, and information systems can all be attacked, but intrusion detection systems can find them. It is challenging to demonstrate the security of the information systems and to uphold such security when they are in use. We put forth the Staked Deep Ensemble Model for Intrusion Detection and Classification (SDEIC) as a solution to these issues. The three stages of feature extraction, optimal feature selection, and classification are used to implement this model. Data normalisation, feature conversion, and feature reduction are all included in the feature extraction phase. A better data normalisation procedure in particular normalises the supplied data. After that, feature conversion is applied to the normalised data. The feature reduction process is then subjected to the Principle Component Analysis (PCA) method. The reduced feature set is then used to select the best features, and this study suggests the Self-Upgraded Squirrel Search Optimisation (SUSSO) algorithm for doing so. In order to classify the specified features, the ensemble model—which consists of the models RNN (Recurrent Neural Network), Improved Deep Belief Network (IDBN), and Deep Max-out Network—is introduced. Additionally, the suggested SUSSO algorithm optimises the weights of the Deep Max-out model during training to guarantee accurate classification. The projected model's accuracy is 97 .5%, which is 12.5%, 2.3%, 41.9%, 1.6%, 18.8% and 15.8% higher than the previous models like NN, RNN, DBN, GRU, and SVM. Lastly, the effectiveness of the suggested approach outperforms the conventional methods. and has obtained satisfactory results.