Applying Survival Analysis for Business Time-to-Event Problems

Description

The course will be presented by Michael J. A. Berry or Gordon S. Linoff , co-founders of Data Miners, Inc. and co-author of Data Mining Techniques for Marketing, Sales and Customer Relationship Management and Mastering Data Mining

This course introduces survival analysis in the context of business data mining. The focus is on understanding customer behaviors that have a time-to-event component. Understanding how long a customer will remain active is the first step in calculating the future value of that customer. Two data characteristics, discrete time and large volume, make it possible to estimate hazards without making restrictive assumptions about the underlying distribution or form or the hazard function. Empirical hazards provide a window on customer behavior that is useful in itself and also provides the basis for calculating survival curves.

Prerequisite Skills

There are no formal prerequisites. A background in business analysis, statistics, or mathematics is helpful but is not essential. The class exercises utilize SAS Enterprise Guide and some programming code is shown. Familiarity with SAS Enterprise Guide or with SAS programming is not required.

SAS Modules Used

This course addresses SAS Enterprise Guide.

Course Topics

Introduction

• course introduction
• history and background of survival analysis

Tools and Data

• visualizing survival
• introduction to data sets used in the class
• introduction to SAS Enterprise Guide

Understanding the World by Counting Point Estimation of Hazards

• why empirical hazards
• visualizing the hazard calculation
• the hazard calculation
• defining the start and the stop
• censoring

Estimating Survival

• tracking survival from one day of starts
• calculating and visualizing survival for all customers
• averaging hazards
• retention versus survival
• quantifying survival

The Role of Covariates

• introduction to covariates
• stratification for time-zero covariates
• Cox proportional hazards regression
• time-zero covariates whose effects change
• creating the right covariates

Competing Risks

• risks compete for customers
• conditional hazards and conditional survival

Left Truncation

• defining left truncation
• fixing left truncation by filtering
• fixing left truncation with time windows

Time Windows

• overview
• changes in survival over time
• seasonal effects and policy changes

Scoring Survival Models and Forecasting

• scoring survival models
• forecasting for new starts
• forecasting for existing customers
• forecasting all customers

Time-Varying Covariates

• time-varying versus time-zero covariates
• the cohort approach
• direct hazard estimation for one-time events
• the Cox proportional hazards approach

Repeating Events

• representing repeating events
• modeling the time to next event
• modeling time to nth event

Beyond Empirical Hazards

• fully parametric models
• semi-parametric models
• flexible hazard models