An Efficient Network Log Anomaly Detection System using Random Projection Dimensionality Reduction

Abstract

Network traffic is increasing all the time and network services are becoming more complex and vulnerable. To protect these networks, intrusion detection systems are used. Signature-based intrusion detection cannot find previously unknown attacks, which is why anomaly detection is needed. However, many new systems are slow and complicated. We propose a log anomaly detection framework which aims to facilitate quick anomaly detection and also provide visualizations of the network traffic structure. The system preprocesses network logs into a numerical data matrix, reduces the dimensionality of this matrix using random projection and uses Mahalanobis distance to find outliers and calculate an anomaly score for each data point. Log lines that are too different are flagged as anomalies. The system is tested with real-world network data, and actual intrusion attempts are found. In addition, visualizations are created to represent the structure of the network data. We also perform computational time evaluation to ensure the performance is feasible. The system is fast, finds real intrusion attempts and does not need clean training data.