This project, a collaboration with the department of Computer Science, is based mainly on data from the VISTA VVV/VVVX surveys, the first large scale infrared exploration of the Milky Way in the time domain (co-led by Dr Lucas). The VVV and VVVX surveys observed about 1 billion stars in a large part of the Milky Way over a 5 year period, detecting several million variable stars in the near infrared. The aim of this project is to develop a machine learning method to classify the variable stars via their light curves, with the principal goal of detecting new types of variable star. No prior knowledge of machine learning is required. The method will be supplemented by more classical statistical approaches (e.g. period search tools) and multiwaveband data on star colours and motions. The datasets have already yielded many unclassifiable high amplitude variable stars and transients, typically very red, optically obscured stars likely having circumstellar matter. This project can be quite flexible given the range of different approaches and questions that can be explored. The student would also undertake observational follow-up of newly discovered unusual variables using telescopes in Chile. We expect that the methods developed can also be applied to other datasets, e.g. the YSOVAR mid-infrared project and many areas of human activity where unevenly sampled time series data are important.