quest q1 2014 - sas dimensional modeling groups/quest...4 step methodology. university of new south...
TRANSCRIPT
Using SAS® Enterprise BI Server 9.2 and
Dimensional Modeling Techniques
to
Identify Students that May Need Support
QUEST Q1-2014
Data
Scientists
?
Agenda
� Introduction to UNSW and the Australian School of Business
� Description of the business problem
� 4 step methodology
University of New South Wales
� UNSW
� Formed in 1949
� More that 50,000 students
� Member of the Group of Eight (Go8)
� Ranked 52 in the QS World University Rankingshttps://www.unsw.edu.au/sites/default/files/documents/UNSW4009_Miniguide_2012_AW2_V2.pdf
University of New South Wales
� Australian School of Business
� Over 12,000 students
� Currently ranked 12th in the world for Accounting and Finance degrees
� Top ranking MBA in Australia
� MBA ranked 48th in the world
http://en.wikipedia.org/wiki/University_of_NSW
Julia Enterprise Data Warehouse
� Developed by the Institutional Analysis and Reporting Office at UNSW with support from UNSW IT
� SAS was installed at UNSW in 2004 as a proof of concept
� 2009 Migrated from SAS 9.1 to 9.2
� 2010 Julia in its current form commenced in SAS Enterprise BI Server using Kimball dimensional modeling techniques
� Flagged for replacement by an EDW being developed by UNSW IT using SAS Enterprise BI Server 9.4
Business Problem� Identification of students potentially at risk
� Widespread, automated and earlier student advisement related to engagement and performance
� Student engagement in courses via Learning Management System (LMS) access and activity
� How do you identify two or three hundred students out of 12000 needing support?
� Students are often shy in asking for help
Methodology
Step 2 –Analyse for Churn or
Risk Patterns
Step 3 – Build a Repeatable
Model
Step 4 –Apply the
Model
Step 1 –Obtain Good
Customer Data
SAS Enterprise Guide®
Star Schemas in SAS BI Suite
SAS BI Suite
Step One: Obtain Good Customer Data /
Build a Good Data Warehouse
Step Two: Analyse for Churn or Risk
Patterns
Convention in the sector
� Low Social Economic Standing
� Low ATAR (Australian Tertiary Admissions Rank)
� Students with a lower WAM (Weighted Average Mark)
Are much more likely to drop out
Step Two: Analyse for Churn or Risk
Patterns
� What is Risk?
� Low WAM
� Churn (Dropping out of UNSW)
� A number of variables were investigated for Churn and WAM using SAS Enterprise Guide
Step Two: Analyse for Churn or Risk
Patterns
� Variables investigated for Churn and WAM using SAS Enterprise Guide� Admittance Type - Cross Institutional, Exchange Student, Foundation Studies UNSW, First Year Student, Internal Program
Transfers, Readmit to Program etc.
� Application Method - Direct or University Admissions Centre
� Social Economic Standing by Postcode – Based on ABS data
� Gender – Retention and WAM comparisons
� Language Spoken at Home
� High School Math - Subject and Grades
� Parental Education Level
� English Language Proficiency for international students
� Residency group – Local or international
� Students in a program that was not their first choice
� Blackboard and Moodle Usage – Learning Management System
� Moodle grades
� Age as the start of program
� Subjects Failed (tested against churn only)
� WAM falling (tested against churn only)
Step Two: Analyse for Churn or Risk
Patterns
Highest Parental Education Level vs. Retention
Step Two: Analyse for Churn or Risk
Patterns
Geo mapping of WAM
Step Three: Build a Repeatable Model� Decided on three groups of attributes:
� Current Learning Activities – Given the most weight� LMS Exam Result Rate� LMS Access Rate
� University Study History – Given the second most weight� Failed This Course Before� Course Fails� WAM Drop Level� WAM Level
� University Entry Ranks – Given the least weight� ATAR Score� High School Math Proficiency� Ranked Entry Score� Written English Proficiency � Total English Proficiency
Step Three: Build a Repeatable Model
� We built a linear model fairly simple, able to be explained (one of the goals)
� Ultimate would be to have multiple models and evolve them over time and potentially select students who show up in the models
� We still don’t KNOW what is happening in the student’s life
Predictive Analytics: The Power to Predict Who Will Click, Buy, Lie, or Die by Eric Siege
Step Four: Apply the Model
� Pilot – picked four subjects and ran a pilot program doing intervention� Showed that the model was helping us find students we need to
talk to
� Allowed focus on building methods for intervening
� Output of model fed into CRM from Semester 2 2013
� 2014 – Beginning to focus on risk for specific courses such as Math intensive course, possible expansion to include Physics
Recap
Step 2 –Analyse for Churn or
Risk Patterns
Step 3 – Build a Repeatable
Model
Step 4 –Apply the
Model
Step 1 –Obtain Good
Customer Data
SAS Enterprise Guide®
SAS Enterprise BI ServerSAS® Data Integration Studio SAS® Web Report Studio
SAS Enterprise BI ServerSAS Data Integration Studio SAS Web Report Studio
Conclusion� Questions?
� Contact
David Waters
Ph: 0408-074082
Linkedin: https://www.linkedin.com/in/davidmwaters