Data Mining and Predictive Modeling

[#1]

See how to:

• Understand the manufacturing yield example used in the demo
• Find patterns
  • Use Distribution to examine the relationship between variables and between variables and response
  • Use Graph Builder to examine all variables, use icon drag-and-drop to fit lines to data, turn on statistics like R square, change fit type interactively and add color to highlight key findings for others
• Perform exploratory modeling
  • Use Multivariate Analysis to see if there is correlation between variables 
  • Examine Scatterplot Matrix and use Predictor Screening to find meaningful relationships
• Deploy JMP Pro predictive modeling to handle non-normal responses and uncover potential overfitting
• Use Generalized Regression to create model and randomly segment data into 3 sets: Training set estimates model parameters, Validation set assesses model's predictive ability by getting best or worse as terms are added, Test set is left out of model-building and gives the final, independent assessment of model predictive ability
• Choose Distribution approach
• Interactively create Validation column for cross-validation to avoid overfitting the model (overestimating predictive ability)
• Screen for multiple models (new in JMP Pro 16) 
• Use Bootstrap Forest and Boosted Trees (Decision Trees) 
• Build a Neural Network model with 2 layers and enough nodes to describe relationship
• Iteratively refine and rebuild best candidate models to improve predictability
• Save model algorithms (score codes) to data table

Note: Q&A begins ~ Time 38:00.

The JMP Journal includes PowerPoint Slides, which are also attached as a .pptx file.  If you are interested in adding PowerPoint to your JMP Journals as slides, here are the instructions: To get slides into journal:

1. Install Brady Brady’s Journal Text Sledgehammer add-in.
2. Generate PowerPoint deck
3. Save as .png (or other graphic format)
4. Open JMP and select Journal Text Sledgehammer from the add-ins menu
5. Click the tab for Utilities
6. Click Slides to Journal
7. Select folder with all slides that was automatically generated in 3 above
8. Choose format (e.g., wide)
9. Save Journal

Resources

• Data Mining Techniques: For Marketing, Sales and Customer Relationship Management by Michael J. A. Berry and Gordon S. Linoff

• JMP documentation on using JMP Projects

• Mastering JMP video and files on predictive modeling using Time Series Forecasting

• JMP Pro Generalized Regression Distribution estimation options mentioned at ~ Time 32:00 in video

• JMP Pro Generalized Regression information. Use NEXT button on documentation pages to read through in sequence.

• Video on Discovering and Predicting Patterns Using Neural Network Models

• Calculations for parameter estimates
• Some comon questions and proposed answers from GitHub.

Questions for the presenter?  Add a comment and mention Jason Higgins using @Jason_Wiggins.

Comments about the value of questions at the end?  Add comment for @gail_massari.

[John_K]

@Jason_Wiggins Hi Jason, great topic, I didn't see the video of the session here -- @gail_massari can you point to the video of Jason's presentation please ?

[Jeff_Perkinson]

@John_K, the video is embedded in the post above. Are you not able to see it? 

[John_K]

@Jeff_Perkinson I see it now, thanks Jeff & Gail