AI-Driven Early Warning System for Predicting Student Dropouts

Abstract

Student dropout is a significant educational issue in India, influenced by regional differences and unequal resources. This paper introduces an AI-driven early warning system for predicting dropout risk at the state level using multi-year UDISE+ data. The proposed framework combines Logistic Regression, XGBoost, and CatBoost through a weighted ensemble strategy to enhance predictive accuracy. It includes automated preprocessing, feature engineering, classification, and time-series forecasting to classify states into Low, Medium, and High risk levels. An interactive Streamlit dashboard allows for visualization and supports decision-making. To connect prediction with action, a Community Learning Hub offers tailored learning and mentoring support for high-risk areas. The modular design allows for scalability and fosters data-driven educational planning.

Introduction

Education plays a crucial role in the socio-economic development of a country. However, student dropout remains a major challenge in India, affecting employability, human capital, and economic growth. Traditional methods based on historical statistics are limited in their ability to predict future dropout risks. Therefore, this research proposes an AI-based early warning system to predict student dropout risks across Indian states using multi-year data from the Unified District Information System for Education Plus (UDISE+).

The proposed system uses educational indicators such as student enrolment, dropout rates, teacher availability, school infrastructure, pupil–teacher ratio, and gender parity index. The data is preprocessed through normalization, feature engineering, and calculation of relevant indicators to improve prediction quality.

The system combines three machine learning algorithms: Logistic Regression, XGBoost, and CatBoost. Their predictions are integrated using a weighted ensemble strategy to improve accuracy and reduce prediction variability. In addition, time-series forecasting techniques such as Prophet are used to analyse historical dropout trends and estimate future risks. The system also incorporates Vidya Setu, a Community Learning Hub, accessible through a Streamlit dashboard, to support educational planning and intervention.

The experimental results compare Logistic Regression, standalone XGBoost, and the proposed CatXGB ensemble model. The ensemble achieves the best reported performance, with 89.19% accuracy, 89.62% precision, 89.19% recall, and an F1-score of 0.8795. The confusion matrix indicates that the model correctly classifies most cases across low, medium, and high dropout-risk categories.

Feature importance analysis identifies total student enrolment and the pupil–teacher ratio as major factors influencing dropout-risk predictions. School infrastructure indicators, including access to electricity and the total number of schools, also contribute to the predictions.

Conclusion

This paper presents an AI-driven early warning system for predicting state-level school dropout risk in India using multi-year UDISE+ data [1]. The framework combines Logistic Regression, XGBoost, and CatBoost through a weighted ensemble to improve prediction accuracy, along with time-series forecasting for trend analysis [8]. The system is implemented using a Streamlit dash-board and integrated with the VidyaSetu Community Learning Hub to support targeted interventions. The experimental results show that the ensemble model performs better than the individual models, proving that it is effective for analysing large-scale educational data and supporting decision-making.

References

[1] Ministry of Education, Government of India provides the Unified District Information System for Education Plus (UDISE+), which can be accessed online at the official website. [2] P. Kumar, S.K. Patel, S. Debbarma, and N. Saggurti (2023) studied the factors that influence school dropout among adolescents in India in their paper published in PLOS ONE. [3] M.N.I. Sarker, M. Wu, and M.A. Hossin (2019) discussed the economic impact of school dropout in Bangladesh in the Journal of Social and Economic Development. [4] K.S. Selim and S.S. Rezk (2023) explored how machine learning can be used to predict school dropout in Egypt, published in Education and Information Technologies. [5] L. Aulck, N. Velagapudi, J. Blumenstock, and J. West (2016) worked on predicting student dropout in higher education, presented at the ICML Workshop on Data for Education. [6] T. Chen and C. Guestrin (2016) introduced XGBoost, a scalable and efficient boosting system, at the ACM SIGKDD Conference. [7] L. Prokhorenkova, G. Gusev, A. Vorobev, A. Dorogush, and A. Gulin (2018) developed CatBoost, a boosting method designed to handle categorical features, presented at NeurIPS. [8] S.J. Taylor and B. Letham (2018) discussed large-scale forecasting methods in their work published in The American Statistician. [9] Streamlit Inc. (2023) provides Streamlit, a framework used for building interactive data applications.

Copyright

Copyright © 2026 Illapu Sankara Srinivasa Rao, Pedumajji Kusuma Priya, Urlapu Anitha Sri, Kurmana Sasank, Pothala Anil Kumar. 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.