Unsupervised Anomaly Detection in Multivariate Time Series Data
Anomaly (outlier) detection is crucially important in a variety of domains such as medical image analysis, fraud detection and spacecraft monitoring. These events could have highly detrimental effects or could cause complete failure of systems that are vital to the business.
There is an exponential increase in multivariate time series data generated by real-world systems and mainly three types of anomalies could be observed: point, contextual and collective anomalies. However, there is no exact definition as these can differ from business to business. Understanding and taking corrective actions become tedious when millions of data points are ingested every second from multiple interrelated systems.
In this talk, Umit will talk about his client experience and various approaches (machine learning and deep learning application) that are used in academia and industry to design, implement and deploy such systems.