Multi-Dimensional Feature-Driven Intrusion Detection Method Based on Improved CNN and LSTM

For enhancing network attack identification performance and addressing the deficiencies inherent in conventional machine learning approaches regarding detection capabilities, this work introduces a network intrusion identification methodology utilizing enhanced Convolutional Neural Network (CNN) combined with improved Long and Short Term Memory Network (HMLSTM). Initially, normalization technology serves to prepare the data, followed by employing Lion Algorithm (LSO) for hyperparameters optimization of CNN, thereby establishing the optimized architecture OCNN, which integrates with HMLSTM model for capturing spatial and temporal features. Subsequently, these spatial-temporal feature vectors enable the training and testing processes of the OCNN-HMLSTM top-level classifier. Multiple widely adopted datasets underwent extensive experimentation in this work. Experimental outcomes demonstrate that substantial enhancement in network intrusion identification accuracy has been achieved, while superior performance emerges concerning detection accuracy alongside false alarm rate when contrasted against alternative methodologies.