Security and Privacy Control Architecture for Outlier Hunting on Network Traffic in Higher Education

Abstract

This research provides a security and privacy control architecture for outlier hunting of internet Network traffic in higher institutions. The architecture leveraged the power of security and privacy combined with deep learning algorithms. By exploiting the hierarchical features learned through multiple layers, the proposed architecture can discern nuanced outliers indicative of security threats, network failures, or performance bottlenecks. The RNN model achieved a lower accuracy of 0.18 compared to the autoencoder. However, it still exhibited high 0.92 precision, 0.91 recall, and 0.91 F1-score values, indicating that it performed well in capturing temporal dependencies in the data. The accuracy of the tuned model (0.81) was significantly higher than that of the base model (0.62), indicating that the hyperparameter tuning and adjustments made to the model architecture led to improved performance in terms of accurately reconstructing the input data, predicting network intrusions and providing security and privacy on network data.