Electroencephalography (EEG) records electrical activity in the brain using a series electrodes placed on the scalp. EEG equipment is relatively cheap and portable and is currently one of the most widely used non-invasive brain imaging tools in neuroscience and in the clinic. In research, EEG recordings are used in a wide range of fields, including medicine (e.g. diagnosis of epilepsy or the early detection of dementia), computer science (e.g. brain computer interfacing ( BCI) and human activity recognition (HAR)) and psychology (e.g. the study of cognitive development). Each field has a range of related tasks and experimental structures, and each has a different default methodology. However, at the heart of many EEG related research questions is the problem of building a predictive regression or classification problem. This can be diagnostic (does the EEG recording of a patient indicate they have dementia?), descriptive (can we tell from the EEG recording whether an individual is moving their left or right arm?), or cognitive (is the subject looking at a picture of a face or random noise?). This project will focus on developing the latest innovations in times series classification [1,2] (TSC) to the general problem of learning from EEG. We will adapt the latest version of the HIVE-COTE algorithm  for EEG classification through embedding spatial information in the transformation process, and collaborate with experts in the field  to compare our novel approaches against accepted gold standard methodologies. We will collate publicly available datasets, develop solutions within an open-source framework  and conduct our own experiments to try and answer the core question of this project: can we automate the process of building predictive models from EEG signals in a way that is as good as, or better than, handcrafted solutions?