High-precision intrusion detection for cybersecurity communications based on multi-scale convolutional neural networks

This study developed an advanced network intrusion detection system based on an improved multi-scale convolutional neural network architecture aimed at enhancing the accuracy of detecting network threats. By precisely capturing data features at different scales, this system significantly improves the model’s ability to analyze complex network behaviors. The proposed system incorporates a novel data preprocessing method combining SMOTE and ENN techniques to address the class imbalance in datasets while resolving the overlap issue of minority and majority class samples present in the SMOTE algorithm. It also utilizes a novel feature selection approach combining Information Gain, Random Forest feature importance scoring, and Recursive Feature Elimination to optimize model performance and reduce computational load. Experiments conducted on public datasets CICIDS2017, KDDCUP99, and UNSW-NB15. The experimental results demonstrate that intrusion detection based on a multi-scale convolutional neural network exhibits high detection accuracy. Specifically, the accuracy on the KDDCUP99 and CICIDS2017 datasets all exceeded 99.85%, while on the UNSW-NB15 dataset surpassed 99.20%, indicating the method’s ability to accurately identify network intrusions.