Improved Parallel Cross-Convolutional Neural Network Based Network Traffic Classification and Anomaly Detection for Real-Time Analysis

Nowadays, the rapid growth of network traffic and increasing sophistication of cyber threats have made it essential to develop efficient and effective network traffic analysis techniques. The existing Convolutional Neural Networks (CNN) have learnt patterns and anomalies in network traffic for real-time classification. However, it has led to challenges including complexity of network traffic patterns and low false positives. Hence, this research proposes an Improved Parallel Cross-CNN (IPC-CNN) to provide better results by learning diverse feature representations with multiple convolutional branches simultaneously. Initially, the input data is acquired from USTC-TFC2016 dataset which contains real- world network traffic payloads including time series characteristics and then fed into preprocessing with normalization as well as one-hot encoding. Here, the normalization helps to standardize the traffic data and one-hot- encoding is used to transform the categorical data into numerical format. Finally, IPC-CNN is employed to detect and classify the network traffic payloads by extracting spatial features through parallel convolutional layers. From the results, the proposed IPC-CNN achieved better results in terms of accuracy (99.21%) when compared with existing Genetic Algorithm-Support Vector Machines (GA-SVM).