About the Project

Inspiration

Student dropout is a serious educational challenge caused by various factors, including poor academic performance, irregular attendance, financial difficulties, limited family support, health concerns, and lack of access to technology. We realized that traditional student information systems mainly store historical records but often fail to identify students who need help before they drop out.

This inspired us to build DropGuard – AI-Powered Student Dropout Prediction System, a proactive solution that helps schools identify students at risk, understand the reasons behind their risk, and take timely action to support them. Our goal is to use artificial intelligence not just to predict problems but to help prevent them.

What it does

DropGuard is an AI-powered platform that predicts student dropout risk and provides actionable insights to help educators make informed decisions.

Key features include:

Early Dropout Prediction: Uses machine learning to classify students into Low, Medium, and High dropout risk levels.
Multi-Dimensional Student Analysis: Evaluates academic performance, attendance, behavior, family support, financial conditions, health indicators, and technology access.
Explainable AI: Highlights the major factors influencing each prediction, helping educators understand why a student may be at risk.
Personalized Intervention Recommendations: Suggests appropriate actions, such as academic tutoring, attendance monitoring, financial assistance, or counselling.
School Analytics Dashboard: Provides visual insights into student statistics, risk distributions, attendance, and trends.
Student Data Import: Supports CSV and Excel imports with validation, column mapping, duplicate detection, and error reporting.
Secure Role-Based Access: Provides separate permissions for system administrators, headmasters, and teachers, with school-level data isolation.
Prediction History and Reports: Tracks changes in student risk levels and enables data exports for further analysis.

How we built it

We developed DropGuard using a modern full-stack architecture that integrates web development, database management, and machine learning.

Frontend: React with Vite, Tailwind CSS, Axios, Framer Motion, and Lucide React for a responsive and interactive user interface.

Backend: Python with FastAPI, SQLAlchemy, Pydantic, and Alembic for API development, data validation, database operations, and schema migrations.

Database: PostgreSQL, with structured tables for schools, students, academic records, attendance, behavior, family information, health, technology access, predictions, and activity logs.

Machine Learning: We implemented a tuned Logistic Regression pipeline using scikit-learn. The model processes 24 predictive input features to estimate dropout probability and classify risk levels. Its coefficients help explain the important factors influencing individual predictions.

Data Processing: We implemented a data-import pipeline that validates CSV and Excel files, maps columns, normalizes input values, and checks for errors before storing student records.

Security: We integrated JWT-based authentication, bcrypt password hashing, role-based access control, school-scoped data access, and activity logging.

The model was trained using a dataset containing 647 student records. During validation, it achieved 85.32% accuracy and a 95.93% ROC-AUC score. These metrics reflect the current evaluation results and may vary on new, unseen student data.

Challenges we ran into

Handling diverse student data: Academic, behavioral, family, health, and technology-related information require different validation and preprocessing methods.
Building a reliable prediction pipeline: We needed to ensure that the same preprocessing steps and feature order were used during training and prediction.
Making predictions understandable: A risk score alone is insufficient, so we used Logistic Regression coefficients to provide interpretable risk-factor insights.
Designing useful recommendations: Connecting identified risk factors to appropriate interventions required clear rules for academic support, attendance, financial assistance, and counselling.
Maintaining data security: Student information is sensitive, so implementing role-based permissions and school-level isolation was an important challenge.
Ensuring reliable data imports: Spreadsheet files may contain missing values, duplicate records, inconsistent formats, or incorrect column names, requiring careful validation and error reporting.

Accomplishments that we're proud of

Built an end-to-end student dropout prediction platform integrating a web application, backend APIs, database, and machine learning.
Developed a predictive model with 85.32% validation accuracy and 95.93% ROC-AUC on the current dataset.
Integrated explainable AI insights to show the major factors contributing to predicted dropout risk.
Implemented targeted intervention recommendations to help educators identify possible support actions.
Designed a school-level dashboard for monitoring student statistics and dropout risk distributions.
Implemented secure, multi-role access and school-level data isolation.
Developed bulk student data import with validation and diagnostic reporting.
Established a foundation for tracking student risk changes through prediction history.

What we learned

Building DropGuard helped us understand how machine learning can be integrated into a practical software application. We learned about data preprocessing, feature engineering, Logistic Regression, model evaluation, and explainable AI.

We also strengthened our knowledge of full-stack development, REST APIs, relational database design, authentication, role-based access control, data validation, and secure handling of sensitive information.

Most importantly, we learned that an AI prediction is only one part of a solution. The real value comes from turning predictions into understandable insights and timely human action. Dropout risk predictions should support educators, not replace their judgment, and should be used carefully to avoid unfairly labelling students.

What's next for AI-Powered Student Dropout Prediction System

Our future plans focus on making DropGuard more useful, scalable, and effective in real educational environments.

Automated Alerts: Notify teachers, counsellors, and guardians when a student requires timely attention.
Intervention Outcome Tracking: Record the support provided to students and monitor whether their risk levels improve.
Longitudinal Analysis: Track student progress and risk trends across multiple academic terms.
Model Retraining: Develop workflows to update the machine learning model as new, appropriately collected data becomes available.
Improved Model Evaluation: Validate performance on larger and more diverse datasets, assess calibration and subgroup fairness, and monitor prediction quality over time.
**Regional