Network Traffic Anomaly Detection Using t-Distributed Stochastic Neighbor Embedding and Variational Autoencoder Approach

The rapid expansion of technology and connected devices has intensified the focus on cyberattacks that compromise devices and generate massive traffic volumes disrupting Internet services. This paper presents a t-Distributed Stochastic Neighbor Embedding and Variational Autoencoder (t-SNE-VAE) for efficient network traffic anomaly detection. The input data was preprocessed by using min-max normalization where it provided standard scaling to avoid imbalance in data. The t-Distributed Stochastic Neighbor Embedding (t-SNE) is employed for feature extraction and dimensionality reduction. The Variational Autoencoder (VAE) with high reconstruction ability is employed to detect anomaly in network traffic. Using NSL-KDD dataset, the proposed method reduces data dimensionality and captures complex patterns with significant improvement in detection accuracy. The proposed approach enhances the detection of network anomalies, improves the interpretability and robustness of the model. The proposed t-SNE-VAE model achieved high accuracy of 99.87%, precision of 99.50%, and recall of 99.34%, surpassing several state-of-the-art methods such as Semi-supervised Gated Recurrent Unit (SEMI-GRU), Stacked-Deep Polynomial Network with Spider Monkey Optimization (SDPN-SMO).