The OEA Chronic Absenteeism Package provides a set of assets which support an education system in developing their own predictive model to address chronic absenteeism. There are two main components of this package:
- Guidance and documentation: The OEA Chronic Absenteeism Package - Use Case Documentation provides guidance on the end-to-end process of developing a successful Chronic Absenteeism use case project, including how to engage stakeholders in the project, prior research on the use case problem domain and theory, how to map data sources to the theory of the problem, and how to implement Microsoft’s Principles of Responsible Data and AI in the process of predictive modelling. It is highly recommended this document be reviewed by any education system considering using this package, and that the documentation be revised to the specific data and decisions for that system’s context.
- Technical assets: Various assets are freely available in this package to help accelerate implementation of Chronic Absenteeism use cases. Assets include descriptions of data sources, notebooks for data processing, a pipeline for ML model building and deployment, and sample PowerBI dashboards. See descriptions of technical assets below.
Important Note: It is strongly recommended to education systems or institutions planning to use this package establish that they establish a process for obtaining student, family, guardian, teacher, faculty, and staff consent for using this type of student absense data. This consent should be part of the system or institution’s broader data governance policy that clearly specifies who can have access to what data, under what conditions, for what purposes, and for what length of time.
This OEA Package was developed through a partnership between Microsoft Education, Kwantum Analytics, and Fresno Unified School District in Fresno, California.
Chronic absenteeism is generally defined as a student missing 10% or more of a school year. Student absenteeism is a fundamental challenge for education systems which has increased as result of the global pandemic. There is a growing body of research (see Use Case Documentation) substantiating what most parents and teachers have long believed to be true: School truancy undermines the growth and development of students. Students with more school absences have lower test scores and grades, a greater chance of dropping out of school, and higher odds of future unemployment. Absent students also exhibit greater behavioral issues, including social disengagement and alienation. The most recent national estimates in the US suggest that approximately 5–7.5 million students, out of a K–12 population of approximately 50 million, are missing at least 1 cumulative month of school days in a given academic year, translating into an aggregate 150–225 million days of instruction lost annually.
Machine learning models offer the potential to find patterns of absenteeism across student attendance patterns, class engagement, academic achievement, demographics, social-emotional measures and more. Predictions of students at risk of becoming chronically absent allows for targeted support of these students. A predictive model can be used to precisely focus resources to support students who are on the trajectory of chronic absenteeism, identify the best interventions to prevent absenteeism, and ultimately reduce absenteeism.
This package was developed in collaboration with Fresno Unified School District in Fresno, California and already has created an impact (see the Use Case Documentation for details).
In general, this package can be used by system or institutional leaders, school, or department leaders, support staff, and educators to:
- accurately identify which students are at risk of becoming chronically absent or may move to a higher tier of absence
- quickly understand what type of support resources or interventions might be most effective to prevent or reduce absenteeism with individual students
- guide decision making of school support staff by providing a real-time and detailed snapshot of students who are at risk of higher level of absence based on engagement, academic, and well-being patterns of that student.
See below for examples of developed PowerBI dashboards.
Explanation Page | Overview of Chronic Absenteeism | Chronic Absenteeism Drivers |
---|---|---|
Patterns of absenteeism | Strongest drivers of model predictions | School support staff dashboard |
---|---|---|
Preparation: Ensure you have proper Azure subscription and credentials and setup v0.6.1 of the OEA framework. This will include v0.6.1 of the OEA python class. Note: This package will be updated to accommodate v0.7.
- Examine available data sources. See below for these related data sources. Choose which modules or data sources to implement.
- This package was developed using the following modules: Contoso SIS, Microsoft Education Insights, and Clever.
- Run each of the module test data pipelines to ingest the test data into stage 2.
- Use the Digital Engagement Schema pipeline and process the compatible modules you choose to ingest. This will combine all digital activity module-tables into a unified table, and creates a single database for the Power BI dashboard. Visit the Data page for a detailed explanation of its use in the PowerBI data model.
- Import and run the Chronic Absenteeism package pipeline template.
- This step will automatically kick-off the development of the StudentModel table, used to train and test the ML model. The ML model is automatically trained and tested in the second notebook.
- In the second notebook, the top 5 model-determined drivers are identified are pushed to a table, as well as: the true chronic absence flag, the predicted chronic absence flag, the model-dedicated probability of chronic absence for that student, and the model certainty of predictions.
- Use the Power BI dashboard to explore predicting Chronic Absenteeism. Note that all pipelines create SQL views which can be accessed via your Synapse workspace serveless SQL endpoint. Example dashboard concepts are provided in this package.
- More detailed information on these queries are provided in the Power BI folder.
The machine learning model learns from past student data to predict if a student will become chronically absent in the future. The model building and assessment is done in 5 main steps:
- Data collection: Select and aggregate data needed to train the model (described below).
- Feature engineering: Use education context to combine and normalize data.
- Model trianing: InterpretML is used to train a model. The best model is used to score the training dataset with predictions.
- Model prediction interpretations: The InterpretML Explainable Boosting Classifier is used to identify which features are most impactful (called key drivers) on the model predictions.
- Fairness and PowerBI: Training data, model predictions, and model explanations are combined with other data such as student demographics. The combined data is made ready for PowerBI consumption. PowerBI enables assessment of model quality, analysis of predictions and key drivers, and analysis of model fairness with respect to student demographics.
- Important Note: This package does not currently incorporate this, due to lack of test data. It is highly recommended the notebooks are updated to incorporate this data, as well, for production purposes.
See the Chronic Absenteeism Package Data page for understanding how to deploy this package using test data, Documentation for details on migrating to production data, and Pipelines for more details on model building.
This package combines multiple data sources which were identified through research as strongly related to absenteeism:
- School Information System (SIS): School, grade, and roster data
- Barriers to students: Transportation data, distance from school, school changes, student illness
- School experiences: School suspension, disciplinary, behavior, and learning outcome data
- Engagement data: School attendance, digital engagement
This package can use several OEA Modules to help ingest data sources that are typically used to understand patterns of chronic absenteeism (see below for list of relevant OEA modules).
OEA Module | Description |
---|---|
Ed-Fi Data Standards | For typical Student Information System (SIS) data, including detailed student attendance, demographic, digital activity, and academic data. |
Microsoft Digital Engagement | Such as M365 Education Insights, or Microsoft Graph data. |
Clever | for learning app data |
i-Ready | for language and math assessments and learning activities. |
These modules are then combined into single tables based on the types of data contained with them, using the OEA schemas to ingest and transform the module data so that only the relevant columns are extracted from the stage 2 data. Below is the list of relevant OEA schema definitions used in this package.
OEA Schema | Description |
---|---|
Digital Engagement Schema | For extracting forms digital engagement into a standardized OEA schema. |
This Predicting Chronic Absenteeism package was developed by Kwantum Analytics in partnership with Fresno Unified School District in Fresno, California. The architecture and reference implementation for all modules is built on Azure Synapse Analytics - with Azure Data Lake Storage as the storage backbone, and Azure Active Directory providing the role-based access control.
Assets in the Chronic Absenteeism package include:
- Data: For understanding the data relationships and standardized schema mappings used for certain groups of data.
- Documentation:
- Use Case Documentation developed with Fresno Unified School District.
- Resources and documentation for Machine Learning in Azure, InterpretML, and Responsible AI implementations.
- Notebooks: For cleaning, processing, and curating data within the data lake.
- Pipelines: For an overarching process used to train the machine learning model and support PowerBI dashboards.
- PowerBI: For exploring, visualizing, and deriving insights from the data.
Microsoft and any contributors grant you a license to the Microsoft documentation and other content in this repository under the Creative Commons Attribution 4.0 International Public License, see the LICENSE file, and grant you a license to any code in the repository under the MIT License, see the LICENSE-CODE file.
Microsoft, Windows, Microsoft Azure and/or other Microsoft products and services referenced in the documentation may be either trademarks or registered trademarks of Microsoft in the United States and/or other countries. The licenses for this project do not grant you rights to use any Microsoft names, logos, or trademarks. Microsoft's general trademark guidelines can be found at http://go.microsoft.com/fwlink/?LinkID=254653.
Privacy information can be found at https://privacy.microsoft.com/en-us/
Microsoft and any contributors reserve all other rights, whether under their respective copyrights, patents, or trademarks, whether by implication, estoppel or otherwise.