/* ---- Google Analytics Code Below */

Saturday, August 06, 2016

Predictive Analytics Interview

Good piece, non technical, relatively little about how results will be integrated with the decision process.

Talk with Eric Siegel on predictive analytics.  Good, non technical interview:

" ... What is predictive analytics? How do you define it, and how did you get interested in working with predictive analytics? 

As defined in my book, “Predictive Analytics,” this is technology that learns from experience (data) to predict the future behavior of individuals in order to drive better decisions.”  ... 

As a result, predictive analytics affects everyone every day. Trendsetters like Chase, Facebook, Google, Hillary for America, HP, IBM, Match.com, Netflix, the NSA, Pfizer, Target, and Uber are seizing upon the power of big data to predict human behavior—including yours. Why? Predictive analytics reinvents industries and runs the world. It combats risk, boosts sales, fortifies healthcare, optimizes social networks, toughens crime fighting, and wins elections.

What have been some of the main advances that have fueled the rise of predictive Analytics in recent years?

Ensemble models (Chapter 6 of my book) and uplift modeling (Chapter 7). They are game changers.
Ensemble models are the most prevalent way to advance from simple methods like decision trees in order to improve predictive performance. As with most advanced method, you lose transparency (model interpretability) — but you don’t lose all that much simplicity: an ensemble model is nothing more than a collection of models such as decision trees that vote (or are otherwise combined). This is kind of like collective intelligence — the wisdom of a crowd not of people but of predictive models. ... 

How is predictive analytics and machine learning related? 

The two are largely synonymous. In academics and R&D labs, it is called “machine learning” (although that is a slightly broader area), and when deployed commercially, it is “predictive analytics.” ... 

This makes sense, since the machine needs to learn in order to predict. It learns from data how to consider the factors (variables) known about an individual in order to predictively score that individual. The mechanism that considers multiple factors to derive a single score, e.g., a decision tree or neural network, is called a predictive model; the process of creating the model from data – modeling – is the learning process, and the model is what’s been learned. ... " 

No comments: