Anomaly Detection In Telemetry Data Using Ensemble Machine Learning

This paper explores unsupervised machine learning methods for anomaly detection in telemetry datasets by reviewing and identifying best-automated detection algorithms and methodologies for anomaly detection. There have been various research to identify an effective model to detect anomalies for telemetry data to reduce response time so as to mitigate risks and avoid failures. Traditional algorithms for anomaly detection have trouble identifying attacks throughout the data analysis task. Machine learning approaches, such as supervised, and unsupervised methods for grouping, classification, and regression, appear to be very useful tools for analyzing anomalous behavior. These techniques can identify any anomalous behavior in telemetry data and allow room for research into the real-time analysis. The principal aim of this research is to answer the question "How can we improve on the current machine learning models for anomaly detection in telemetry datasets?". The dataset consists of five Time-Series datasets and is representative of the data with which we are concerned. Five algorithms are applied to these datasets and examined in depth. Then, three unsupervised anomaly definitions are examined.