An Intelligent Flask-Based Web Application for Early Diabetes Risk Prediction

Abstract

Early detection of diabetes can substantially reduce long-term complications and healthcare costs, yet traditional diagnostic pathways remain resource-intensive and in- accessible to many populations. In this work, we present a comprehensive design, implementation, and assessment of a Flask-based web application that leverages a stacked ensemble using artificial intelligence models to forecast individual diabetes risk using routine clinical parameters. The platform integrates secure user authentication, personalized trend visualizations, and an administrative dashboard for population-level analytics. We detail our methodology—from data preprocessing and feature engineering to model training and web deployment—evaluate system performance on benchmark and real-world datasets, and discuss the broader implications for scalable preventive healthcare solutions.

Introduction

Diabetes mellitus is a chronic metabolic disorder marked by persistent high blood sugar due to insulin issues, affecting over 400 million people globally with serious complications if unmanaged. Early detection and risk management are vital to reduce health burdens.

Machine learning (ML) techniques are increasingly used for diabetes prediction. Logistic regression provides interpretability but struggles with complex data, while ensemble methods like Random Forests and Gradient Boosting improve predictive accuracy. Deep learning offers advanced pattern recognition but poses challenges in explainability and deployment. Explainability tools (e.g., SHAP, LIME) help make models more transparent and trustworthy. Existing ML systems often lack integration of real-time prediction with trend analysis and secure access, which this work addresses through a robust ML pipeline and user-centered web platform.

The system uses a modular, three-tier architecture: a responsive frontend (HTML5, Bootstrap, Chart.js), a Flask-based backend managing API endpoints and sessions, and a scalable database layer with SQLAlchemy and PostgreSQL. Data flows from user input through preprocessing and ensemble ML inference to storage and visualization, supporting extensibility and interoperability with health records.

Data was sourced from the PIMA dataset and a local cohort, with preprocessing steps including missing value imputation, outlier removal, and feature engineering (interaction terms, clustering, PCA). A stacked ensemble combining logistic regression, Random Forest, and XGBoost was trained and optimized, achieving strong performance (AUC-ROC 0.89, accuracy 82%) and robustness to class imbalance.

The web application backend uses Flask with secure session and form handling, while the frontend emphasizes usability and interactive visualizations. Comprehensive testing—including unit, integration, system, load, and security tests—ensured accuracy, reliability, and protection against vulnerabilities. A pilot user study confirmed high usability and trust in the system’s explainability.

Overall, the project delivers a scalable, secure, and interpretable diabetes risk prediction platform that integrates advanced ML methods with a user-friendly web interface, providing a strong foundation for future enhancements and deployment.

Conclusion

This essay offers a scalable, secure, and user-friendly web application for early diabetes risk prediction. By combining interpretable ML models with interactive visual analytics, our platform empowers individual users and healthcare administrators to make data-driven decisions. Future enhancements will focus on: 1) Automated Alerts: SMS/email notifications for high-risk assessments. 2) Wearable Device Integration: Real-time data ingestion from glucose sensors. 3) Continuous Learning: Automated retraining pipelines using new user data. 4) Microservices Migration: Containerization and orchestration for enterprise-scale deployment.

References

[1] E. Kavakiotis et al., “Prediction of type 2 diabetes using machine learning techniques,” Comput. Struct. Biotechnol. J., vol. 15, pp. 104–111, 2017. [2] T. Chen and C. Guestrin, “XGBoost: A scalable tree boosting system,” in Proc. 22nd ACM SIGKDD, 2016, pp. 785–794. [3] S. Lundberg and S.-I. Lee, “A unified approach to interpreting model predictions,” in Adv. Neural Inf. Process. Syst., 2017, pp. 4765–4774. [4] M. Grinberg, Flask Web Development, 2nd ed., O’Reilly Media, 2018.

Copyright

Copyright © 2025 Mrs. Sahana S, Dr. S D N Hayath Ali. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.