Predictive Models
Overview
WebAdMIT Predictive Insights uses your institution’s historical applicant and matriculation data to estimate the likelihood that current applicants will enroll if admitted. Predictive models are developed specifically for your institution using historical data from WebAdMIT and are updated annually to incorporate the most recent enrollment outcomes.
Once your model is implemented, predictive insights are delivered directly to WebAdMIT and updated daily as applicant information changes throughout the admissions cycle.
For information about accessing, interpreting, and using predictive scores in WebAdMIT, review Accessing and Using Predictive Scores.
How Predictive Models Are Developed
Liaison's data science team works with your institution to develop a predictive model using historical applicant data from WebAdMIT.
The model development process generally includes:
- Collecting historical data: Liaison securely collects the WebAdMIT data needed to develop your institution’s predictive model. Predictive models generally require three years of historical data to identify enrollment patterns.
- Reviewing and validating data: Liaison works with your institution to validate application and matriculation counts for the applicable historical cycles and confirm that the data represents the intended applicant population.
- Developing the model: After the data is validated, Liaison uses the historical data to develop your institution’s predictive model.
- Reviewing model performance: Your institution receives information about model performance and high-level insights into the features that influence predictions.
- Integrating predictions with WebAdMIT: Once the model is finalized, predictions and related data are delivered directly to WebAdMIT for use within your existing workflows.
Data Used in Predictive Models
Predictive models use a standard set of applicant information available through WebAdMIT. Depending on your CAS and available data, this can include:
- Applicant demographics
- Academic background, such as institutions attended and GPAs
- Test scores
- Experiences and activities
- Application information
- Geographic information
- Program-specific information
Specific data elements and configurations can vary by CAS and institution.
Keeping Predictions Current
After your predictive model is implemented, WebAdMIT data is sent to the model each day. The model uses the latest available applicant information to generate updated predictions, which are then returned to WebAdMIT.
As a result, an applicant’s predictive insights can change throughout the admissions cycle as new applicants enter the pool or applicant information changes.
Annual Model Updates
Predictive models are retrained annually to incorporate the most recent matriculation outcomes. This allows the model to account for the latest enrollment data and changes to relevant CAS structures over time.