A novel deep learning and FAOA-based DXCov-FH net for intrusion detection in cloud computing using MapReduce framework
TL;DR
A novel method, named Deep Xception convolutional Forward Harmonic Network with Frigate Ablation Optimization Algorithm (DXCov-FH Net_FAOA), which exhibited high detection performance and effectiveness for intrusion detection, confirming its effectiveness for intrusion detection.
Abstract
Intrusion Detection in Cloud Computing focuses on analyzing large amount of log data and identifying security threats. Existing approaches struggle with false-positive reduction, scalability, real-time monitoring, and adapting to new attack techniques. To address these issues, a novel method, named Deep Xception convolutional Forward Harmonic Network with Frigate Ablation Optimization Algorithm (DXCov-FH Net_FAOA), is developed. Initially, the cloud environment is modeled to establish network security. The MapReduce framework is subsequently employed to process the generated log data through its Mapper and Reducer components. Weitendorf’s Linear (WL) technique is employed within the Mapper phase for input data normalization. Next, feature selection is carried out using Gini Impurity-based Weighted Random Forest (GIWRF), which evaluates features according to impurity and weighting criteria. while Synthetic Minority Over-sampling Technique (SMOTE) is subsequently applied to increase the representation of the available data. Intrusions are identified in the Reducer phase through DXCov-FH Net, a hybrid approach that incorporates XCovNet, Harmonic Analysis, and Deep Stacked Autoencoder (DSA) techniques. FAOA is employed to optimize the model during training by combining the principles of Magnificent Frigatebird Optimization (MFO) with the Snow Ablation Optimizer (SAO). DXCov-FH Net_FAOA exhibited high detection performance, recording 97.10% for TNR, 97.99% for TPR, 96.57% for precision, and 97.27% for F1-score. The model further achieved 97.59% accuracy and a ROC-AUC of 0.999. The reported performance was further supported by 95% confidence interval analysis, demonstrating the reliability of the obtained results. Experimental results indicate that the proposed approach provides accuracy enhancements of 6.83%, 4.86%, 4.64%, 1.99%, 1.37%, and 0.96% over existing techniques, confirming its effectiveness for intrusion detection.