In business, data and algorithms create economic value when they reduce uncertainty about financially important outcomes. This course teaches the concepts and mathematical methods behind the most powerful and universal metrics used by data scientists to evaluate the uncertainty-reduction—or information gain—predictive models provide.

We focus on the two most common types of predictive model—binary classification and linear regression—and you will learn metrics to quantify for yourself the exact reduction in uncertainty each can offer. These metrics are applicable to any form of model that uses new information to improve predictions cast in the form of a known probability distribution—the standard way of representing forecasts in data science.

In addition, you will learn proper methodology to avoid common data-analytic pitfalls when forecasting—such as being “fooled by randomness” and over-fitting “noise” as if it were “signal.”

Course Offered by Coursera — Duke University