Introduction
Introduction¶
Predictive modelling and machine learning have recently become increasingly important in biomedical research and hold promise for delivering biomarkers that substantially impact clinical practice and public health [1, 2, 3, 4]. For many researchers, “machine learning” is a mysterious term that may be more familiar from science-fiction than their actual research activity. At the same time many of these researchers routinely apply rather complex statistical techniques like mixed-effect models and find their way in highly complex analysis workflows of dedicated software environments. .
These type of researchers are the primary target readership of this lecture book. The aim of this book is to demystify predictive modelling and machine learning and to show that the simplest - but yet very powerful - forms of these analysis techniques can as much accessible and straightforward as “classical” tools for statistical inference, they routinely use.
This book aims to provide a practical hands-on experience and illustrates all the basic concepts by analyzing and example dataset with python code. It invites the reader to an exciting journey, towards predicting age from brain structure. While the example dataset is a structural MRI dataset, the jupyter book does not assume any neuroimaging-related background knowledge. If, nevertheless, the reader is interested in some more background information, the are many good resources available on-line for free. The practice pages of the book are interactive notebooks: they can be run either on a local workstation or - with just one click - in the cloud. The book contains exercises which require the reader to modify the code. It is encouraged to go through the book with an own dataset, instead of the example dataset. This approach may be although more challenging, but it will make the reader deeply internalize the core concepts and - hopefully - push her over the entry limit for applying predictive modelling in her own research. While this lecture book focuses on simple and easily explainable machine learning models, it also highlights, how the paradigm of predictive modelling makes it possible to use “black box” model, without having to sacrifice confidence in the reliability of the predictions.
This is not a conventional lecture book. This is an interactive jupyter book, being constantly developed. Please contribute to making it better by leaving feedback or opening an issue at the github repository of the book.
Have a lot of fun while discovering a new way of dealing with your data.