DAT-650: Advanced Data Analytics

Case Study, Part 1: Business Understanding, Data Preparation, and EDA

• Cross-Industry Standard Practice for Data Mining
• Business Understanding Phase
• Data Understanding Phase, Part 1: Getting a Feel for the Data Set
• Data Preparation Phase
• Data Understanding Phase, Part 2: Exploratory Data Analysis

Milestone

Multivariate Statistics

• Two-Sample t-Test for Difference in Means
• Two-Sample Z-Test for Difference in Proportions
• Test for the Homogeneity of Proportions
• Chi-Square Test for Goodness of Fit of Multinomial Data
• Analysis of Variance
• Reference
• The R Zone
• R Reference
• Exercises

Case Study, Part 2: Clustering and Principal Components Analysis

• Partitioning the Data
• Developing the Principal Components
• Validating the Principal Components
• Profiling the Principal Components
• Choosing the Optimal Number of Clusters Using Birch Clustering
• Choosing the Optimal Number of Clusters Using k-Means Clustering
• Application of k-Means Clustering
• Validating the Clusters
• Profiling the Clusters

k-Nearest Neighbor Algorithm

• Classification Task
• k-Nearest Neighbor Algorithm
• Distance Function
• Combination Function
• Quantifying Attribute Relevance: Stretching the Axes
• Database Considerations
• k-Nearest Neighbor Algorithm for Estimation and Prediction
• Choosing k
• Application of k-Nearest Neighbor Algorithm Using IBM/SPSS Modeler
• The R Zone
• R References
• Exercises

Association Rules

• Affinity Analysis and Market Basket Analysis
• Support, Confidence, Frequent Itemsets, and the A Priori Property
• How Does The A Priori Algorithm Work (Part 1)? Generating Frequent Itemsets
• How Does The A Priori Algorithm Work (Part 2)? Generating Association Rules
• Extension From Flag Data to General Categorical Data
• Information-Theoretic Approach: Generalized Rule Induction Method
• Association Rules are Easy to do Badly
• How Can We Measure the Usefulness of Association Rules?
• Do Association Rules Represent Supervised or Unsupervised Learning?
• Local Patterns Versus Global Models
• The R Zone
• R References
• Exercises

Multiple Regression and Model Building

• An Example of Multiple Regression
• The Population Multiple Regression Equation
• Inference in Multiple Regression
• Regression With Categorical Predictors, Using Indicator Variables
• Adjusting R2: Penalizing Models For Including Predictors That Are Not Useful
• Sequential Sums of Squares
• Multicollinearity
• Variable Selection Methods
• Gas Mileage Data Set
• An Application of Variable Selection Methods
• Using the Principal Components as Predictors in Multiple Regression
• The R Zone
• R References
• Exercises

Variable Selection Methods

NaïVe Bayes and Bayesian Networks

• Bayesian Approach
• Maximum A Posteriori (MAP) Classification
• Posterior Odds Ratio
• Balancing The Data
• Naïve Bayes Classification
• Interpreting The Log Posterior Odds Ratio
• Zero-Cell Problem
• Numeric Predictors for Naïve Bayes Classification
• WEKA: Hands-on Analysis Using Naïve Bayes
• Bayesian Belief Networks
• Clothing Purchase Example
• Using The Bayesian Network to Find Probabilities
• The R Zone
• R References
• Exercises

Imputation of Missing Data

• Need for Imputation of Missing Data
• Imputation of Missing Data: Continuous Variables
• Standard Error of the Imputation
• Imputation of Missing Data: Categorical Variables
• Handling Patterns in Missingness
• Reference
• The R Zone
• R References

Case Study, Part 3: Modeling And Evaluation For Performance And Interpretability

• Do You Prefer The Best Model Performance, Or A Combination Of Performance And Interpretability?
• Modeling And Evaluation Overview
• Cost-Benefit Analysis Using Data-Driven Costs
• Variables to be Input To The Models
• Establishing The Baseline Model Performance
• Models That Use Misclassification Costs
• Models That Need Rebalancing as a Surrogate for Misclassification Costs
• Combining Models Using Voting and Propensity Averaging
• Interpreting The Most Profitable Model

Case Study, Part 4: Modeling and Evaluation for High Performance Only

• Variables to be Input to the Models
• Models that use Misclassification Costs
• Models that Need Rebalancing as a Surrogate for Misclassification Costs
• Combining Models using Voting and Propensity Averaging
• Lessons Learned
• Conclusions