Machine learning (ML) can provide unique analytical insights, as well as help automate some operational and decision-making processes more efficiently and effectively than non-ML alternatives. However, ML is also among the buzziest of buzzwords, and many are overselling and oversimplifying its usage.

Do not let anyone frame a data analysis, business problem, or process improvement as an ML use case. Instead, say: That is Not Machine Learning — that is a data analysis, business problem, or process improvement where ML might be able to help. But not before we evaluate other options. And with the understanding that ML is rarely going to be either the first or only aspect of the solution.

This episode is sponsored by: Vertica.com

Extended Show Notes: ocdqblog.com/dbp

Follow Jim Harris on Twitter: @ocdqblog

Email Jim Harris: ocdqblog.com/contact

Other ways to listen: bit.ly/listen-dbp

Label Making. That is my simple two-word definition of Machine Learning. Machine Learning is Label Making. ML is LM.

Especially supervised machine learning, which creates either numerical labels (using regression algorithms) to make predictions about a continuous data value (such as sale or stock prices), or categorical labels (using classification algorithms) to assign data to pre-defined groups also called classes (such as Fraud or Not Fraud for financial transactions).

This episode is sponsored by: Vertica.com

Extended Show Notes: ocdqblog.com/dbp

Follow Jim Harris on Twitter: @ocdqblog

Email Jim Harris: ocdqblog.com/contact

Other ways to listen: bit.ly/listen-dbp

Based on one of my presentations, this episode provides a five-part vendor-neutral framework for evaluating the critical capabilities of a cloud data analytics solution: Deploy, Store, Optimize, Analyze, Govern.

This episode is sponsored by: Vertica.com

Extended Show Notes: ocdqblog.com/dbp

Follow Jim Harris on Twitter: @ocdqblog

Email Jim Harris: ocdqblog.com/contact

Other ways to listen: bit.ly/listen-dbp

A decade ago, just before the beginning of the data science hype cycle was the big data hype cycle. At that time I had the privilege of sitting down with Ph.D. Statistician Dr. Thomas C. Redman (aka the “Data Doc”).

We discussed whether data quality matters less in larger data sets, if statistical outliers represent business insights or data quality issues, statistical sampling errors versus measurement calibration errors, mistaking signal for noise (i.e., good data for bad data), and whether or not the principles and practices of true “data scientists” will truly be embraced by an organization’s business leaders.

This episode is an edited and slightly shortened version of that discussion, which even though it is from ten years ago, I think it still provides good insight into big data quality, then and now.

Extended Show Notes: ocdqblog.com/dbp

Follow Jim Harris on Twitter: @ocdqblog

Email Jim Harris: ocdqblog.com/contact

Other ways to listen: bit.ly/listen-dbp

Before you get started on any data analytics effort, you need to have at least preliminary answers to three questions: (1) What problem are we trying to solve?, (2) What data can we apply to that problem?, and (3) What analytical techniques can we apply to that data?

This episode is sponsored by: Vertica.com

Extended Show Notes: ocdqblog.com/dbp

Follow Jim Harris on Twitter: @ocdqblog

Email Jim Harris: ocdqblog.com/contact

Other ways to listen: bit.ly/listen-dbp

In time for opening day of the 2022 Major League Baseball (MLB) season, I discuss the initial results of my Baseball Data Analysis Challenge.

See the extended show notes for links to my input data, my results as a Microsoft Excel file, and my SQL scripts on GitHub.

I used logistic regression machine learning classification models to calculate win probabilities for the Boston Red Sox across nine (9) game metrics, and a Naïve Bayes machine learning classification model to predict individual game wins and losses with an associated probability.

Think you can best my model? Game on! The baseball data analysis challenge continues. Play ball!

Extended Show Notes: ocdqblog.com/dbp

Follow Jim Harris on Twitter: @ocdqblog

Email Jim Harris: ocdqblog.com/contact

Other ways to listen: bit.ly/listen-dbp

Why don’t more machine learning models graduate to production? Paige Roberts stops by to help explore this topic and drop some knowledge about how to get more machine learning models deployed in production.

This episode is sponsored by: Vertica.com

Extended Show Notes: ocdqblog.com/dbp

Follow Jim Harris on Twitter: @ocdqblog

Email Jim Harris: ocdqblog.com/contact

Other ways to listen: bit.ly/listen-dbp

Back in 2012, Harvard Business Review declared Data Scientist was The Sexiest Job of the 21st Century. Less than a year later, I recorded a podcast discussion with an actual data scientist and Ph.D. Statistician, Dr. Melinda Thielbar, during which she discussed what a data scientist actually does and provided a straightforward explanation of key concepts, such as signal-to-noise ratio, how statistical results should be presented and explained to various audiences, uncertainty, predictability, experimentation, and correlation.

This episode is an edited and slightly shortened version of that discussion, which even though it is from nine years ago, I think it still provides good insight into data science, then and now.

Extended Show Notes: ocdqblog.com/dbp

Follow Jim Harris on Twitter: @ocdqblog

Email Jim Harris: ocdqblog.com/contact

Other ways to listen: bit.ly/listen-dbp