Machine learning-based detection of anomalous behaviour on streaming time series multivariate data is under active research today. It has potential in several domains, such as network performance monitoring for abnormal network behaviour detection. However, most existing anomaly detection approaches require manual intervention and extensive domain knowledge to tune the detection engine appropriately. Moreover, they provide an offline detection engine rather than dealing with real-time anomaly detection online.

The candidate’s assignment is to develop and implement a machine learning-based real-time anomaly detection engine for streaming multivariate time series or modify an existing offline detector for online detection on real-time streaming multivariate data. Thus, in the frame of the bachelor thesis work, the following tasks have to be completed:

· Investigate state-of-the-art multivariate anomaly detection approaches; 

· Design and implement a machine learning-based real-time anomaly detection engine for streaming multivariate time series or modify an existing offline detector for online detection on real-time streaming multivariate data; 

· Test and validate the developed solution;

· Compare the performance of the developed solution to other state-of-the-art multivariate anomaly detection approaches;

· Examine and discuss the possibilities for further improvements.