An Efficient and Intelligent Deep Learning Techniques for Detection and Classification of Intrusion Detection System in Class Imbalance Datasets

Abstract

Malicious code development greatly hinders the creation of Intrusion Detection Sys- newlinetems (IDS). Because malware developers utilize various information-concealing eva- newlinesion tactics to resist identification by an intrusion detection system, these attacks be- newlinecome complicated, making it particularly hard to recognize obfuscated and unknown newlinemalware. When it comes to cyber security, an IDS is essential for spotting harmful newlineactions in network traffic. Nonetheless, the disparity in class sizes has led to a diffi- newlinecult problem where there are more students in some classes than others. Traditional newlineclassifiers, therefore, struggle to categorize malicious activity and have little resistance against unidentified bugs. newlineThe first paper suggests using an Imbalanced Generative Adversarial Network(IGAN) newlineto increase the efficiency of class discrimination while addressing the issue of class imbalance. We standardized the raw data and used one-hot encoding during data Pre- newlineprocessing to reduce the impact of extreme values and outliers on the overall results. newlineLong Short-Term Memory(LSTM) and Lenet 5 are utilized as an ensemble to categorize newlineevents that are deemed anomalous into different attack categories. The outcomes of the inquiry demonstrate that the proposed strategy works higher than the other deep learning techniques, reaching better precision, accuracy, recall, F1-score, TPR, and FPR. newlineThe statistics showed that the recommended approach valued performance measures newlinesignificantly more than competing methods. In comparison to existing classifiers, the newlineproposed methodology is found to classify different attacks with an accuracy of above newline98%. newlineOur second approach, the Attention-based Nested U-Net (ANU-Net), utilizes deep newlinelearning techniques to address these issues and enhance the performance of intrusion newlinedetection. The initial data preprocessing for this IDS model is completed in three steps: newlineduplication removal, label transformation, and data normalization. The features are newlinesubsequently retrieved and chosen in accordance with th