An Introduction to Machine Learning Interpretability

More about the book

Innovation and competition are pushing analysts and data scientists towards complex predictive modeling and machine learning algorithms. While these models enhance accuracy, they also complicate understanding, leading to diminished human trust, which can hinder business adoption, regulatory oversight, and model documentation. Sectors like banking, insurance, and healthcare particularly require interpretable predictive models. This ebook by Patrick Hall and Navdeep Gill from H2O.ai introduces machine learning interpretability and explores various techniques, algorithms, and models that enable data scientists to enhance predictive accuracy while preserving interpretability. Readers will learn about the application of machine learning in practice, the social and commercial motivations behind interpretability, fairness, accountability, and transparency. The text also differentiates between linear models and more complex machine learning models, provides a definition of interpretability, and highlights key research groups in the field. Additionally, it offers a taxonomy for classifying interpretable machine learning approaches, practical techniques for data visualization, training interpretable models, and generating explanations for complex predictions, along with automated methods for testing model interpretability.