This article presents a solution for predicting student attrition. Core components include:
- Azure Machine Learning for training and deploying machine learning models.
- Responsible AI Toolbox for identifying contributing factors and detecting bias.
- Azure Synapse Analytics for data processing.
Download a Visio file of this architecture.
Data source. Educational data comes from various sources:
- District records
- Digital archives of instructional materials and grade books
- Detailed information about financial grants
- Student responses to course surveys
Educational data is retrieved from various data sources, including district databases and state databases. It's housed in an on-premises database.
Data preparation. During data preparation, the data is gathered, combined, structured, and organized. It's then ready to be used:
- To build machine learning models.
- For business intelligence purposes.
- In analytics and data visualization applications.
The solution uses Azure Data Factory to orchestrate the process of transforming and loading the data. It uses Azure Synapse Analytics to process data and to trigger Machine Learning experiments.
AI machine learning training. The solution uses Azure Machine Learning studio to train a wide range of supervised learning algorithms and to find a model that accurately predicts student attrition. The following Responsible AI Toolbox tools help to implement responsible AI:
- The interactive interpretability dashboard helps users understand the major factors that contribute to student attrition.
- The fairness tool provides an interactive dashboard for detecting and mitigating bias that's related to student gender and race in the chosen model.
AI machine learning inferencing. During inferencing, previously unseen data points are fed into a machine learning model. The model calculates the probability of student attrition. In Machine Learning, a built-in model registry stores and provides version control for models in the Azure cloud. The model registry makes it easy to organize and keep track of trained models. Trained models are deployed to instances of Azure Data Science Virtual Machine or managed endpoints.
Analytical workload. The model scoring results are stored in Azure Synapse Analytics and Azure SQL Database. The results are then available for use in the front end and for monitoring and retraining the models.
Front-end model consumption. Power BI and the Web Apps feature of Azure App Service consume the scored results.
Student data schema
The information that's critical for the student attrition model consists of factors that influence student behavior. The following table lists data elements that affect student attrition and retention patterns. The list isn't complete.
|Gender||The gender that the student was assigned at birth or "Not reported."|
|Race||The student's reported race: either "Black," "White," "Pacific Islander," "Asian," "American Indian," or "Not reported."|
|First generation in college||Whether the student is a first-generation student. With a first-generation student, no parent or guardian has a four-year degree or higher from a college or university.|
|Total terms||The total number of terms that the student has been enrolled.|
|High school graduate or GED||Whether the student graduated from high school or has an equivalent education.|
|Cumulative grade point average (GPA)||The average over all the grades that the student has earned.|
|Cumulative credit hours earned||All the hours that the student has accumulated over the period of enrollment.|
|Transferred to another program||Whether the student has switched programs since starting.|
|Financial aid eligible||Whether the student is eligible for a grant.|
|Age||Minimum||The minimum age that's recorded for the period of enrollment.|
|Age||Maximum||The maximum age that's recorded for the period of enrollment.|
|Entry type||Dual enrollment||Whether the student is enrolled in a dual program.|
|Entry type||Early admission||Whether the student was admitted in advance of the usual notification date.|
|Entry type||First time in college||Whether the student is enrolled in college for the first time.|
|Entry type||Other||Other admission types, indicated as a numerical value.|
|Entry type||Re-entry||A numerical value that represents re-entry admission.|
|Entry type||Transfer||A numerical value that represents transfer admission.|
|Academic standing||Academic probation||The period in which the student must improve academic standing, expressed as a normalized value in proportion to the enrollment period.|
|Academic standing||Academic suspension||Whether the student has been suspended, expressed as a normalized value in proportion to the enrollment period.|
|Academic standing||Academic suspension for one year||The proportion of the enrollment period that the student has been suspended for one academic year.|
|Academic standing||Academic warning||Whether the student has been given an academic warning for failing grades.|
|Academic standing||Extended probation for low GPA||The period that the student's probation has been extended in proportion to the enrollment period.|
|Academic standing||Good academic standing||The proportion of the enrollment period that the student has been in good academic standing.|
|Academic standing||Probation after suspension or dismissal||Whether the student has been placed on probation after academic suspension.|
|Instruction type||Blended||The proportion of time that the student has received a combination of in-person and online instruction.|
|Instruction type||Fully online||The proportion of time that the student has received only online instruction.|
|Instruction type||Remote learning||Whether the student has enrolled in remote learning.|
|Instruction type||Remote learning blended||The proportion of time that the student has combined in-person instruction with remote learning.|
|Instruction type||Traditional||The proportion of time that the student has received conventional in-person instruction.|
|Faculty type||Faculty||The proportion of instruction that the student has received from academic staff at the institution.|
|Faculty type||Adjunct||The proportion of instruction that the student has received from contract professors.|
|Faculty type||Unknown instructor type||The proportion of instruction that the student has received of unknown type.|
|Attrition||Whether the student has dropped out of college.|
Azure Data Lake offers limitless storage for data in different shapes and formats. Besides enterprise-grade security and monitoring support, Azure Data Lake integrates easily with Azure analytics tools. Built on top of Azure Blob Storage, Azure Data Lake can manage large amounts of unstructured data, such as archives and data lakes. The service is a good fit for high-performance computing, machine learning, and cloud-native workloads. This solution provides a local data store for the machine learning data and a premium data cache for training the machine learning model.
SQL Database is a fully managed database engine for modern cloud applications. This database service offers built-in intelligent optimization, global scalability and availability, advanced security options, and dynamic scalability with no downtime. SQL Database can automatically process relational data and non-relational structures such as graphs and JSON, spatial, and XML data. For this service's availability guarantee, see SLA for Azure SQL Database.
Data Factory is an orchestration and cloud extract-transform-load (ETL) tool. Besides offering over 90 built-in connectors across various data sources, Data Factory provides copy and transformation functionality in a no-code environment. You can use its diagram view to monitor and manage data integration processes.
Azure Synapse Analytics is an analytics service for enterprise data warehouses. This service uses SQL and Spark technologies and offers dedicated or serverless options for querying data. Azure Synapse Analytics provides a unified experience for ingesting, exploring, preparing, transforming, and managing data. The service also makes data available for business intelligence and machine learning purposes.
Machine Learning is an enterprise-grade machine learning service for easy model development and deployment to a wide range of machine learning targets. This service provides users at all skill levels with a low-code designer, automated machine learning, and a hosted Jupyter notebook environment that supports various integrated development environments (IDEs).
Azure Machine Learning studio is a cloud service that you can use to accelerate and manage the machine learning project lifecycle. This service handles training, model deployment, and the management of machine learning operations (MLOps).
Responsible AI Toolbox is a collection of integrated tools and functionalities that help you implement responsible AI principles. The toolbox integrates ideas from several open-source tools in the areas of error analysis, interpretability, fairness, counterfactual analysis, and causal decision-making. You can use this open-source framework to assess machine learning models quickly and easily.
Azure Data Science Virtual Machine is a customized virtual machine (VM) image on the Azure cloud platform that's built specifically for data science. The image has many popular data science tools preinstalled and preconfigured to jump-start building intelligent applications for advanced analytics.
Machine Learning endpoints are HTTPS endpoints that clients can access to retrieve inferencing output, or scoring output, from a trained model. Each endpoint provides a stable scoring URI with key-token authentication.
App Service provides a framework for building, deploying, and scaling web apps. The Web Apps feature is an HTTP-based service for hosting web applications, REST APIs, and mobile back ends. With Web Apps, you can develop in .NET, .NET Core, Java, Ruby, Node.js, PHP, or Python. Applications easily run and scale in Windows and Linux-based environments.
Power BI is the Azure software as a service (SaaS) for business analytics and visually immersive and interactive insights. Power BI provides a rich set of connectors to various data sources, easy transformation capabilities, and sophisticated visualization capabilities.
- This solution uses Machine Learning as a data modeling and deployment tool. Instead, you can use Azure Databricks with a code-first approach.
- The data sources in the solution are Azure components. You can also use training data from third-party sources.
Student attrition is one of the most common challenges faced by educational institutions across the globe. Attrition is prevalent at all levels of both private and public education systems, leading to various consequences. For example, in the public system, student attrition results in ineffective use of tax revenue. At private institutions, attrition damages reputations and negatively affects school revenue. When student attrition levels are elevated, fewer graduating professionals enter the workforce, which threatens the global economy.
Given these consequences, educational institutions are heavily investing in ways to reduce student attrition. It has become increasingly important for these institutions to identify and address the root causes of attrition. Understanding the major contributing factors helps institutions support individual students and implement at-scale solutions to reduce attrition rates. Institutions also must be able to predict attrition rates to minimize attrition across a broad set of program offerings.
Predicting student attrition is a binary classification problem that predicts whether a student will leave school. This type of model is built on student-centric data that includes demographic data, financial data, the student's academic history, and the course delivery mode.
For interpretability and bias detection, this solution uses the Responsible AI Toolbox. When educational institutions understand the factors that impact continuous education, they can take appropriate actions to curtail student attrition. But institutions need to use models with no bias toward protected classes like gender and race. This point is important, as is the need to consider and implement responsible AI practices. Equitable models help provide unbiased support for student success.
Potential use cases
This solution applies to many areas:
- Adaptive learning. In education, adaptive learning is crucial for student success. After taking into account an individual student's progress, educational institutions can support educators and customize approaches to provide the best possible learning experience for the student.
- Employee attrition prediction. Employees are valuable assets of any organization. It's important to know whether employees are dissatisfied, or whether there are other reasons employees might leave their jobs. When employers have this information, they can take proactive measures to retain employees.
- Customer churn prediction. In retail, churn prediction helps identify whether users are likely to stop using a website, service, or product. Companies and large corporations suffer losses when they can't retain customers. The customer churn model helps these organizations identify ways to improve their offerings and prevent customers from leaving them.
These considerations implement the pillars of the Azure Well-Architected Framework, a set of guiding tenets that you can use to improve the quality of a workload. For more information, see Microsoft Azure Well-Architected Framework.
The technologies in this solution were chosen for their scalability and availability, with the aim of managing and controlling the costs.
Reliability ensures your application can meet the commitments you make to your customers. For more information, see Overview of the reliability pillar.
The components in this solution feature high availability. But there are two parts to machine learning and analytics tasks: training and production deployment. Resources that are required for training don't typically need high availability. As for production deployment, Azure VMs fully support high availability. For detailed information, see Availability options for Azure Virtual Machines.
Security provides assurances against deliberate attacks and the abuse of your valuable data and systems. For more information, see Overview of the security pillar.
The components in this solution offer built-in security. They also support permissions that you can manage by using Azure Active Directory (Azure AD) authentication or role-based access control. For information about establishing suitable enterprise-level security, see Azure Machine Learning best practices for enterprise security.
Azure Synapse Analytics features enterprise-grade and industry-leading security that uses component isolation to protect data, secure networks, and boost threat protection. Component isolation can minimize exposure to threats during security attacks. Azure Synapse Analytics also offers data obfuscation to protect sensitive personal student data.
SQL Database provides security capabilities at all levels, from data protection and data masking to threat protection. For more information, see An overview of Azure SQL Database and SQL Managed Instance security capabilities.
When you implement security features in this solution, consider guidelines in the following resources:
- Deploy dedicated Azure services into virtual networks
- Enterprise security and governance for Azure Machine Learning
One aspect of cost optimization is looking at ways to reduce unnecessary expenses and improve operational efficiencies. For more information about creating a cost-effective workload, see Overview of the cost optimization pillar.
- To optimize costs by paying only for what you need, scale resources according to your analytics, training, and deployment workloads.
- To estimate the cost of implementing this solution, use the Azure Pricing Calculator.
- For information about various licenses that Power BI offers, see Power BI pricing.
Performance efficiency is the ability of your workload to scale to meet the demands placed on it by users in an efficient manner. For more information, see Performance efficiency pillar overview.
You can scale most components in this scenario up or down depending on the analysis activity levels. Azure Synapse Analytics provides scalability and high performance. At low activity levels, you can pause this service or scale back computing resources.
You can scale Machine Learning according to the size of your data and the compute resources that you need for model training. For deployment, you can scale compute resources based on your expected load, scoring service, and latency requirements with Azure Kubernetes Service (AKS).
For guidance about designing scalable solutions, see Performance efficiency checklist.
Follow MLOps guidelines to standardize and manage an end-to-end machine learning lifecycle that's scalable across multiple workspaces. Before you go into production, ensure that the implemented solution supports ongoing inference with retraining cycles and the automated redeployment of models. For more information, see the following resources:
As part of Machine Learning, responsible AI is based on the six pillars of AI use and development:
- Reliability and safety
- Privacy and security
For an overview of responsible AI and detailed implementation information, see What is responsible AI?.
This article is maintained by Microsoft. It was originally written by the following contributors.
- Manasa Ramalinga | Senior Cloud Solution Architect, US National CSA Team
- Charitha Basani | Senior Cloud Solution Architect, US National CSA Team
- Angela Kunanbaeva | Senior Cloud Solution Architect, US National CSA Team
To see non-public LinkedIn profiles, sign in to LinkedIn.
- What is Azure Machine Learning?
- Track ML models with MLflow and Azure Machine Learning
- What is automated machine learning (AutoML)?
- What is Azure Synapse Analytics?
- Unleash the power of predictive analytics in Azure Synapse Analytics with machine learning and AI
- Introduction to Azure Data Lake Storage Gen2
- Advanced analytics architecture
- What is Power BI?
- What is Azure Active Directory?
- About Azure Key Vault