Hybrid Multiscale Residual Features For Network Anomaly Detection

This study proposes a novel hybrid multiscale residual feature approach for network traffic anomaly detection. Utilizing the NSL-KDD dataset, we employed random forest feature selection to identify the 10 most efficient feature sets for classification. These features were then classified using a hybrid deep learning model incorporating multiscale residual blocks and Bidirectional Long Short-Term Memory (Bi-LSTM) layers. The performance evaluation revealed a significant improvement in accuracy compared to previous state-of-the-art methods. Specifically, the proposed approach achieved an accuracy of 99.19% for binary classification and 96.67% for multiclass classification. The lower performance in multiclass classification highlights the challenges posed by unbalanced and insufficient data for certain network anomaly classes within the dataset. Our findings suggest that this hybrid model provides a robust and effective solution for network anomaly detection. We recommend future research to conduct additional experiments using this approach on other network anomaly datasets to further validate its efficacy and robustness across diverse network anomaly classes. This study contributes to advancing the field of network security by presenting a highly accurate and innovative method for detecting network traffic anomalies.