An Introduction to Machine Learning Interpretability

✍ Scribed by Navdeep Gill, Patrick Hall

• Publisher: O’Reilly Media, Inc.
• Year: 2018
• Tongue: English
• Leaves: 45
• Category: Library

✦ Synopsis

Innovation and competition are driving analysts and data scientists toward increasingly complex predictive modeling and machine learning algorithms. This complexity makes these models accurate but also makes their predictions difficult to understand. When accuracy outpaces interpretability, human trust suffers, affecting business adoption, regulatory oversight, and model documentation.

Banking, insurance, and healthcare in particular require predictive models that are interpretable. In this ebook, Patrick Hall and Navdeep Gill from H2O.ai thoroughly introduce the idea of machine learning interpretability and examine a set of machine learning techniques, algorithms, and models to help data scientists improve the accuracy of their predictive models while maintaining interpretability.

Learn how machine learning and predictive modeling are applied in practice
Understand social and commercial motivations for machine learning interpretability, fairness, accountability, and transparency
Explore the differences between linear models and more accurate machine learning models
Get a definition of interpretability and learn about the groups leading interpretability research
Examine a taxonomy for classifying and describing interpretable machine learning approaches
Learn several practical techniques for data visualization, training interpretable machine learning models, and generating explanations for complex model predictions
Explore automated approaches for testing model interpretability

✦ Table of Contents

An Introduction to Machine Learning Interpretability Understanding and trusting models and their results is a hallmark of good science. Scientists, engineers, physicians, researchers, and humans in general have the need to understand and trust models and modeling results that affect their work and their lives. However, the forces of innovation and competition are now driving analysts and data scientists to try ever-more complex predictive modeling and machine learning algorithms. Such algorithms for machine learning include gradient-boosted ensembles (GBM), artificial neural networks (ANN), and random forests, among many others. Many machine learning algorithms have been labeled “black box” models because of their inscrutable inner-workings. What makes these models accurate is what makes their predictions difficult to understand: they are very complex. This is a fundamental trade-off. These algorithms are typically more accurate for predicting nonlinear, faint, or rare phenomena. Unfor......Page 7
Machine Learning and Predictive Modeling in Practice......Page 8
Social and Commercial Motivations for Machine Learning Interpretability......Page 9
1-1......Page 11
Figure 1-3......Page 13
Defining Interpretability......Page 15
A Machine Learning Interpretability Taxonomy for Applied Practitioners......Page 16
A Scale for Interpretability......Page 17
Global and Local Interpretability......Page 19
Model-Agnostic and Model-Specific Interpretability......Page 20
Understanding and Trust......Page 21
Common Interpretability Techniques......Page 22
Seeing and Understanding Your Data......Page 23
1-3......Page 26
1-8......Page 29
Sensitivity Analysis: Testing Models for Stability and Trustworthiness......Page 38
Testing Interpretability......Page 40
Machine Learning Interpretability in Action......Page 41
7......Page 42
15......Page 43
1-2......Page 12
Figure 1-4......Page 14
Figure 1-5......Page 24
1-7......Page 28
Figure 1-7......Page 33
1-6......Page 30
1-12......Page 34
1-16......Page 36