[Video introduction] Regression, decision trees, neural networks—along with many other supervised learning techniques—provide powerful predictive insights. Once built, the models can produce key indicators to optimize the allocation of organizational resources.
New users of these established techniques are often impressed with how easy it all seems to be. Modeling software to build these models is widely available but often results in disappointing results. Many fail to even recognize that proper problem definition was the problem. They likely conclude that the data was not capable of better performance.
The deployment phase includes proper model interpretation and looking for clues that the model will perform well on unseen data. Although the predictive power of these machine-learning models can be very impressive, there is no benefit unless they inform value-focused actions. Models must be deployed in an automated fashion to continually support decision-making for residual impact. The instructor will show how to interpret supervised models with an eye toward decisioning automation.
In this half-day seminar, Keith McCormick will overview the two most important and foundational techniques in supervised machine learning, and explain why 70-80% or more of everyday problems faced in established industries can be addressed with one particular machine learning strategy. The focus will be on highly practical techniques for maximizing your results whether you are brand new to predictive analytics or you’ve made some attempts but have been disappointed in the results so far. Veteran users of these techniques will also benefit because a comparison will be made between these traditional techniques and some features of newer techniques. We will explore that while tempting, the newer techniques are rarely the best fit except in a handful of niche application areas that many organizations will not face (at least not in the short term). Participants will leave with specific ideas to apply to their current and future projects.
- When to apply supervised or unsupervised modeling methods
- Options for inserting machine learning into the decision making of your organization
- How to use multiple models for value estimation and classification
- How to properly prepare data for different kinds of supervised models
- Interpret model coefficients and output to translate across platforms and languages, including the widely used Predictive Modeling Markup Language (PMML)
- Explore the pros and cons of “black box” models including ensembles
- How data preparation must be automated in parallel with the model if deployment is to succeed
- Compare model accuracy scores to model propensity scores that drive decisions at deployment.
Who is it for?
- Analytic Practitioners
- Data Scientists
- IT Professionals
- Technology Planners
- Consultants; Business Analysts
- Analytic Project Leaders.
1. How to choose the best machine learning strategy
- How supervised learning compares to other options
- The reality and the hype regarding machine learning
- What are the classic traditional machine learning techniques?
- The two main types of supervised machine learning
2. Decision Trees: Still the best choice for many everyday challenges
- Exploring and interpreting insights with a completed decision tree model
- A brief primer on the various types of decision tree algorithms
- Strategic considerations and advantages of decision trees
- Deployment and bringing you ML models into production
3. Introducing the CART decision tree
- CART under the hood
- Processing various variable types with CART
- Understanding pruning
- How CART handles missing data with “surrogates”
- The “Roshoman effect” in machine learning
4. Additional Supervised Techniques
- Comparing linear regression to neural networks
- How to embrace the benefits of neural networks without actually using them
- Regression trees: how to use decision trees to address regression problems
- What are “ensemble” methods and why are they so popular?
- Keeping your solutions practical and transparent