Machine learning basic concepts for the movement disorders specialist

In this contribution, we provide a basic introduction to key concepts of Machine Learning (ML). ML can be considered the subfield of Artificial intelligence (AI) which concerns the extraction of information from example data in order to enable computer systems to learn specific tasks and improve their performance. We first introduce most important ML paradigms: unsupervised and supervised learning, which can be associated with typical tasks such as clustering and dimensionality reduction or classification and regression, respectively. Key concepts are illustrated in terms of two imaginary “toy” data sets: the first one represents the properties of a set of various vegetables and fruit. The second one relates to data that could be acquired from a cohort of subjects in the context of movement disorders. For the sake of clarity, emphasis is put on concepts of supervised learning, in particular on classification problems. Examples of important ML methods are illustrated in terms of the above-mentioned imaginary data sets. Without going into technical details, we introduce popular approaches such as linear classifiers, neural networks and prototype-based systems. Moreover, the typical practical workflow of ML, including training, validation, and working phase is outlined. We conclude by briefly discussing issues of practical relevance such as the interpretability of ML models and Explainable AI (XAI). Being far from providing a complete overview of the very broad area of Machine Learning, this chapter should provide a starting point for the further exploration of this very active field of research and its increasing importance in medical data analysis