Machine Learning and AI Foundations: Prediction, Causation, and Statistical Inference [Repost]

In the world of data science, machine learning and statistics are often lumped together, but they serve different purposes, and being versed in one doesn’t mean expertise in the other. In fact, applying a statistical approach to a machine learning problem, or vice versa, can lead to confusion more than elucidation.

In this course, Keith McCormick covers how stats and machine learning are different, when to use each one, and how to use all the tools at your disposal to be clear and persuasive when you share your results. He covers topics like: Why correlation is insufficient evidence of causation; the difference between experimental and observational data; and the differences between traditional statistics and Bayesian statistics. Keith also looks at causality, a tricky topic when it comes to using statistics and machine learning to prove something causes something else. If you build machine learning models, run statistical analyses—or especially if you do both, this course is for you.

Learning objectives

• Explain the significance of a p-value for hypothesis testing.
• Define causation, explain the difference between correlation and causation, and illustrate how to demonstrate causation.
• Explain how to detect multicollinearity and describe a strategy for dealing with it.
• Define induction, deduction, falsification, and counterfactual, and then illustrate their significance in model evaluation.
• Explain the diminishing utility of p-values with an increasing number of model parameters.
• Explain how to test model performance in data mining.
• Describe the conflicting goals and philosophy of statistics and data mining.