BIG STEP CONVOLUTION AND ATTENTION MECHANISMS FOR EFFICIENT ABNORMAL TRAFFIC DETECTION

The quality of service and the security of the network are both determined by the identification of anomalous traffic. The primary challenges in aberrant traffic identification, which are caused by feature similarity and the singular dimension of the detection model, are resolved by a big-step convolution neural network traffic detection model that is based on attention. The network traffic properties are assessed, and the raw data is preprocessed and mapped into a two-dimensional grayscale image. Histogram equalization is employed to generate multi-channel grayscale images. An attention method is employed to improve local features by assigning varying weights to traffic characteristics. Ultimately, pooling-free convolution neural networks are employed to extract traffic characteristics at various depths, thereby addressing the deficiencies of convolution neural networks, such as overfitting and local feature omission. The simulation experiment employed both a balanced public data set and genuine data collection. In comparison to SVM, the proposed model is assessed against ANN, CNN, RF, Bayes, and the two most recent models. A 99.5% accuracy rate was achieved through the use of numerous classes in an experimental setting. The proposed model is exceptional in its ability to detect anomalies. The proposed method outperforms current methods in F1, recall, and accuracy. It has been demonstrated that the model is robust in the face of a variety of challenging conditions and is effective in the detection of items.