Scalable Methods for Time Series Classification

Time series classification is a specialised area of machine learning focused on understanding and exploiting dynamic processes such as environmental, patient, and equipment monitoring, and financial markets. However, many of the most accurate methods for time series classification require significant computational resources. This thesis presents three new methods, together representing a significant advance in terms of accuracy versus computational cost. These new methods can process large quantities of time series data in minutes or hours compared to days or weeks for existing methods, allowing us to learn from larger quantities of time series data with lower computational cost.