2. Answering one Critical Question Now
2
• What is the likelihood that a student will achieve a C or better grade?
• What is likelihood that a student will attend class next week?
3. Retention Rates in Australia
3
• University of Divinity 95.5%
• XXXX University 73%
4. Student Success Return on Investment Calculator
Current Full Time Student Numbers 25,000
Income per Student per annum $13,566*
Current Retention Rate 95.00%
Target Retention Rate 96.00%
Increased Revenue $6,783,000
Note: Assumes two years of lost revenue per unsuccessful student
*Derived from Higher Education Research, Facts and Figures November 2015
4
6. 6
Predictive modeling overview
Problem
What is the question
you are trying
to answer?
Ingest
Aggregate the data
from multiple sources
Modeling
Turn the data
into valuable info
(answer the question)
Distribute
Put the valuable info
in the hands of those
who can act on it
7. 7
Blackboard Predict
How it works
• Aggregate student activity data
• Present risk report to faculty/advisors
Predictive Model
• Current class activity from LMS
• Past student information from SIS
• Can include data from other sources: Portfolio,
ebook/content, response systems
LMS/
SIS
Alerts
& reports
Activity
Data
analysis
1
2
4
3
8. 8
Sample model output
Features and weights
Week Zero Model Weekly Model
.00 .05 .10 .15
Current GPA
Course add weeks prior
Transfer credits
EFC
Percent of classes passed
Ethnicity
Class load
Academic year
Course modality
.00 .10 .20 .30 .40
Earned over attempted
Number of page views
Earned over possible
Percent of classes passed
Days since last page view
Posts to faculty
9. 9
Who could intervene?
Instructor
Channel
LMS – visualizations
Message
Provide triage/assistance
to the students who are at risk
Student
Channel
LMS – visualizations
Message
Compare to average; project grades
Advisor/Counselor
Channel
CRM – tasks
or
Portal – visualizations
Message
These are your students who need
the most help
Fairly simplistic approach to processing data.
MAJOR POINT: This solution is about intervention. The system will help to identify who is at risk. It’s up to the institution to act on it.
We can’t make faculty/advisor intervene
Thought experiment – if I gave you a list of 50 students who I KNOW are going to drop out next week, what would you do with that list? If the school doesn’t have a good answer to this question, the product might not be a good fit.
Talk about the kind of data we get from the SIS
And from the LMS
Other sources can be used if they are “rich” enough (if there is enough usage). Usually, SIS and LMS are sufficient
Sample from a past client
Week Zero model is SIS data only. Notice the balanced contribution of features
Weekly model is an aggregation of Week 1 – Week 16 models. Notice how current course grade (earned over attempted) outweighs other features
Model is custom built for the institution based on their historical data. We build one model for each week (Week zero through next-to-last week)
Match the message to the recipient
We are aware of application fatigue. Don’t want to make the user log in to another system
If they are already in the LMS/CRM, then put the info there…in front of them