Detecting Anomalies in Network Time Series Data Using Restricted Boltzmann Machines

Anomaly detection in temporal data sets is crucial in various applications in various fields.Anomalies can be gradual or abrupt, rising, falling, up or down, or irregular and significant Such anomalies have to be identified to ensure the systems do not have compromise on integrity, safety, and dependability.However, conventional anomaly detection techniques prove ineffective where the nature of data is rich and dynamic. To overcome these challenges, Restricted Boltzmann Machines (RBM) method have been used for time series anomaly detection. Due to the capability of encoding complex structures of the time series data, the RBMs can identify anomalies. The RBM method achieves $\mathbf{9 3 \%}$ accuracy. The anomalies are also clustered using k-means clustering technique and feature correlation matrix for network traffic data is determined. The results show that the proposed method outperformed the existing methods.