SAS Institute Of Management Studies - Predictive Modeling with Logistic Regression using SAS 

Meet the Instructor

What You Learn in This Course

Learner Prerequisites

Using Forums and Getting Help

Access SAS Software for this Course

Set Up Data for This Course (REQUIRED)

About the Demos and Practices in this Course

Overview

Introduction

Goals of Predictive Modeling

Terms for Elements in Predictive Modeling

Basic Steps of Predictive Modeling

Applications of Predictive Modeling

Demonstration Scenario: Target Marketing for a Bank

Demo: Examining the Code for Generating Descriptive Statistics and Frequency Tables

Introduction

Data Challenges

Analytical Challenges

Separate Sampling

Avoiding the Optimism Bias: Honest Assessment

Splitting the Data for Model Training and Assessment

Demo: Splitting the Data

Summary

Practice: Exploring the Bank Data for the Target Marketing Project

Practice: Exploring the Veterans' Organization Data Used in the Practices

Question 1.01

Question 1.02

Question 1.03

Practice: Splitting the Data

Overview

Introduction

Understanding the Logistic Regression Model

Constraining the Posterior Probability Using the Logit Transformation

Understanding the Fitted Surface

Interpreting the Model by Calculating the Odds Ratio

Understanding Logistic Discrimination

Estimating Unknown Parameters Using Maximum Likelihood Estimation

Interpreting Concordant, Discordant, and Tied Pairs

Using PROC LOGISTIC to Fit Logistic Regression Models

Demo: Fitting a Basic Logistic Regression Model, Part 1

Demo: Fitting a Basic Logistic Regression Model, Part 2

Scoring New Cases

Demo: Scoring New Cases

Introduction

Understanding the Effect of Oversampling

Understanding the Offset

Demo: Correcting for Oversampling

Summary

Question 2.01

Question 2.02

Practice: Fitting a Logistic Regression Model

Fitting the Model Review

Overview

Introduction

Reasons for Missing Data

Complete Case Analysis

Methods for Imputing Missing Values

Missing Value Imputation with Missing Value Indicator Variables

Demo: Imputing Missing Values

Cluster Imputation

Introduction

Problems Caused by Categorical Inputs

Solutions to Problems Caused by Categorical Inputs

Linking to Other Data Sets

Collapsing Categories by Thresholding

Collapsing Categories by Using Greenacre's Method

Demo: Collapsing the Levels of a Nominal Input, Part 1

Demo: Collapsing the Levels of a Nominal Input, Part 2

Replacing Categorical Levels by Using Smoothed Weight-of-Evidence Coding

Demo: Computing the Smoothed Weight of Evidence

Introduction

Problem of Redundancy

Variable Clustering Method

Understanding Principal Components

Divisive Clustering

PROC VARCLUS Syntax

Selecting a Representative Variable from Each Cluster

Demo: Reducing Redundancy by Clustering Variables

Question 3.01

Practice: Imputing Missing Values

Question 3.02

Question 3.03

Question 3.04

Practice: Collapsing the Levels of a Nominal Input

Practice: Computing the Smoothed Weight of Evidence

Question 3.05

Practice: Reducing Redundancy by Clustering Variables

Introduction

Detecting Nonlinear Relationships

Demo: Performing Variable Screening, Part 1

Demo: Performing Variable Screening, Part 2

Univariate Binning and Smoothing

Demo: Creating Empirical Logit Plots

Remedies for Nonlinear Relationships

Demo: Accommodating a Nonlinear Relationship, Part 1

Demo: Accommodating a Nonlinear Relationship, Part 2

Introduction

Specifying a Subset Selection Method in PROC LOGISTIC

Best-Subsets Selection

Stepwise Selection

Backward Elimination

Scalability of the Subset Selection Methods in PROC LOGISTIC

Detecting Interactions

BIC-based Significance Level

Demo: Detecting Interactions

Demo: Using Backward Elimination to Subset the Variables

Demo: Displaying Odds Ratios for Variables Involved in Interactions

Demo: Creating an Interaction Plot

Demo: Using the Best-Subsets Selection Method

Demo: Using Fit Statistics to Select a Model

Summary of Preparing the Input Variables, Parts 1 and 2

Question 3.06

Practice: Performing Variable Screening

Practice: Creating Empirical Logit Plots

Question 3.07

Question 3.08

Question 3.09

Practice: Using Forward Selection to Detect Interactions

Question 3.10

Practice: Using Backward Elimination to Subset the Variables

Question 3.11

Practice: Using Fit Statistics to Select a Model

Preparing the Input Variables Review

Overview

Introduction

Fit versus Complexity

Assessing Models when Target Event Data Is Rare

Demo: Preparing the Validation Data

Introduction

Understanding the Confusion Matrix

Measuring Performance across Cutoffs by Using the ROC Curve

Choosing Depth by Using the Gains Chart

Effects of Oversampled Data on Performance Measures

Adjusting a Confusion Matrix for Oversampling

Demo: Measuring Model Performance Based on Commonly-Used Metrics

Introduction

Understanding the Effect of Cutoffs on Confusion Matrices

Understanding the Profit Matrix

Choosing the Optimal Cutoff by Using the Profit Matrix

Using the Central Cutoff

Using Profit to Assess Fit

Calculating Sampling Weights

Demo: Using a Profit Matrix to Measure Model Performance

Introduction

Plotting Class Separation

Assessing Overall Predictive Power

Demo: Using the K-S Statistic to Measure Model Performance

Introduction

Comparing ROC Curves of Several Models"

Demo: Comparing ROC Curves to Measure Model Performance

Using Macros to Compare Many Models

Demo: Comparing and Evaluating Many Models, Part 1

Demo: Comparing and Evaluating Many Models, Part 2

Summary

Question 4.01

Question 4.02

Question 4.03

Practice: Assessing Model Performance

Question 4.04

Question 4.05

Question 4.06

Question 4.07

Measuring Model Performance Review

About the Certification Exam