Multi-Channel Gaussian Process Regression for Network Traffic Anomaly Detection via Inter-Channel Correlation

Network traffic anomaly detection is critical for maintaining secure and reliable communications. This paper proposes a Multi-Output Gaussian Process (MOGP) regression approach to detect network traffic anomalies by jointly modeling multiple channels and exploiting their inter-channel correlations. The Gaussian Process Regression (GPR) model provides a predictive mean and uncertainty for each channel, and we leverage this uncertainty to identify anomalies as deviations with low model probability. An inter-channel fusion logic is introduced to distinguish true anomalies from benign correlated fluctuations across channels, reducing false alarms. We evaluate the framework on a synthetic multi-channel network traffic dataset with injected anomalies. Experimental results show that the MOGP-based approach outperforms a single-channel GPR baseline. These results underscore the effectiveness of leveraging inter-channel correlation and uncertainty estimation in network anomaly detection, yielding more robust and accurate identification of abnormal traffic patterns.