Data Science Foundations: Data Assessment for Predictive Modeling

Introduction

• Why data assessment is critical
• A note about the exercise files

1. What Is Data Assessment?

• Clarifying how data understanding differs from data visualization
• Introducing the critical data understanding phase of CRISP-DM
• Data assessment in CRISP-DM alternatives: The IBM ASUM-DM and Microsoft TDSP
• Navigating the transition from business understanding to data understanding
• How to organize your work with the four data understanding tasks

2. Collect Initial Data

• Considerations in gathering the relevant data
• A strategy for processing data sources
• Getting creative about data sources
• How to envision a proper flat file
• Anticipating data integration

3. First Look at the Data

• Reviewing basic concepts in the level of measurement
• What is dummy coding?
• Expanding our definition of level of measurement
• Taking an initial look at possible key variables
• Dealing with duplicate IDs and transactional data
• How many potential variables (columns) will I have?
• How to deal with high-order multiple nominals
• Challenge: Identifying the level of measurement
• Solution: Identifying the level of measurement

4. Data Loading and Unit of Analysis

• Introducing the KNIME Analytics Platform
• Tips and tricks to consider during data loading
• Unit analysis decisions
• Challenge: What should the row be?
• Solution: What should the row be?

5. Describe Data

• How to uncover the gross properties of the data
• Researching the dataset
• Tips and tricks using simple aggregation commands
• A simple strategy for organizing your work

6. Data Description Case Studies

• Describe data demo using the UCI heart dataset
• Challenge: Practice describe data with the UCI heart dataset
• Solution: Practice describe data with the UCI heart dataset

7. Explore Data Basics

• The explore data task
• How to be effective doing univariate analysis and data visualization
• Anscombe's quartet
• The Data Explorer node feature in KNIME
• How to navigate borderline cases of variable type
• How to be effective in doing bivariate data visualization
• Challenge: Producing bivariate visualizations for case study 1
• Solution: Producing bivariate visualizations for case study 1

8. Explore Data Tips and Tricks

• How to utilize an SME's time effectively
• Techniques for working with the top predictors
• Advice for weak predictors
• Tips and tricks when searching for quirks in your data
• Learning when to discard rows
• Introducing ggplot2
• Orientating to R's ggplot2 for powerful multivariate data visualizations
• Challenge: Producing multivariate visualizations for case study 1
• Solution: Producing multivariate visualizations for case study 1

9. Verify Data Quality

• Exploring your missing data options
• Why you lose rows to listwise deletion
• Investigating the provenance of the missing data

10. Missing Data Case Study

• Introducing the KDD Cup 1998 data
• What is the pattern of missing data in your data?
• Is the missing data worth saving?
• Assessing imputation as a potential solution

11. Explore and Verify Case Studies

• Exploring and verifying data quality with the UCI heart dataset
• Challenge: Quantifying missing data with the UCI heart dataset
• Solution: Quantifying missing data with the UCI heart dataset

12. Making the Transition to Data Preparation

• Why formal reports are important
• Creating a data prep to-do list
• How to prepare for eventual deployment

Conclusion

• Next steps