Statistical Prediction and Machine Learning

This book sets out to join two traditions that are often taught separately: classical statistics and modern machine learning. It presents them as parts of a single framework for data science, combining the older model-based approach with newer data-driven methods and linking statistical theory to machine learning algorithms. The authors are an experienced statistics educator and two data scientists, and the book is meant to close the gap between the two fields for readers who already have a basic grounding in statistics and want to gain real command of machine learning. The opening chapter works through illustrative examples, and the second covers the fundamentals of refined optimization. From there the book moves to widely used supervised learning methods: regression, classification, support vector machines, tree algorithms, and range regressions. Unsupervised learning methods are discussed next, followed by a chapter on statistical learning from data that arrives from several sources, either one after another or all at once. A distinctive strength is its broad treatment of statistical learning in medical settings. Real-life applications include hypertension, stroke, diabetes, thrombolysis, and the efficacy of aspirin. The material draws on the authors' own teaching, research, and consulting work in data analytics. Its examples and companion materials put the weight on understanding the data, interpreting results accurately, and uncovering the hidden assumptions behind different methods. The book also points toward possible future methodological developments in data science.