Abstract:

Accurate student enrollment forecasting is a strategic component of educational management, particularly in large-scale public programs responsible for distributing learning materials. In Brazil, the National Textbook Program (Programa Nacional do Livro Did´ atico – PNLD) depends on reliable estimates of the number of students enrolled in basic education to ensure the adequate production and delivery of didactic resources nationwide. Inaccurate predictions can lead either to oversupply, resulting in unnecessary public expenses, or to shortages that compromise students’ access to essential learning materials. Despite the importance of this task, advanced computational techniques for enrollment forecasting remain limited in national policy applications.

This study applies Fuzzy Time Series (FTS) models to predict student enrollment in Brazilian basic education using real datasets from public schools. Although fuzzy-based approaches have gained increasing attention in the fields of educational analytics and dropout prediction, their systematic adoption for enrollment forecasting in large-scale public programs is still limited. We evaluate several FTS variants, including the Chen, Cheng, Ismail, Seasonal FTS and Ensemble-based FTS models, and compare their performance with established statistical forecasting techniques such as ARIMA, Holt–Winters, ETS and SARIMAX. Quantitatively, the FTS models showed lower forecasting errors, with the best-performing variants (e.g., Ensemble FTS and the Seasonal Adaptive Differential FTS Ensemble) reaching a MAE of approximately 12.3–12.4 in the main evaluation horizon, outperforming ARIMA (MAE ≈ 12.5), ETS (MAE ≈ 13.8) and substantially surpassing SARIMAX, which exhibited higher instability in this context.

These results indicate that fuzzy approaches better capture uncertainty and nonlinear fluctuations inherent to educational time series. In addition, the study offers evidence of how computational intelligence can support large-scale decision-making processes and enhance the efficiency of public policies.

By improving the accuracy of enrollment predictions, this research contributes to optimizing material allocation within the PNLD, reducing financial waste, and ensuring equitable access to educational resources. Ultimately, the study reinforces the value of data-driven strategies in strengthening educational planning and positively impacting millions of students across Brazil.

Keywords:

Fuzzy Time Series, Enrollment Forecasting, Educational Planning, National Textbook Pro- gram (PNLD), Time Series Analysis, ARIMA, SARIMAX, Computational Intelligence, Public Education Policy, Resource Allocation.