Abstract:

Accurate enrollment forecasting is a cornerstone of educational planning, supporting decisions on resource allocation, infrastructure investment, and teacher workforce management. In Brazil, this challenge is particularly consequential in the context of the Programa Nacional do Livro e do Material Didático (PNLD), a large-scale public program that distributes textbooks and educational materials to millions of students annually. PNLD procurement cycles require enrollment projections spanning multiple years ahead, making forecast reliability across extended horizons a practical necessity rather than a purely academic concern.

This study investigates the forecasting horizon problem applied to school enrollment series using Fuzzy Time Series (FTS) models. Drawing on a large-scale national dataset of Brazilian basic education schools, we evaluate how predictive accuracy evolves as the forecast horizon extends from one to four years ahead, with training fixed on a pre-evaluation historical window. Multiple FTS approaches are benchmarked against statistical baselines to assess not only absolute performance but the rate of accuracy degradation over time.

Beyond methodological comparison, the study characterizes the practical limits of FTS-based enrollment forecasting: identifying the horizon range within which fuzzy models remain competitive and the point at which alternative approaches should be considered. Results are discussed in terms of their implications for PNLD planning cycles, offering guidance for practitioners seeking to align forecasting tool selection with the temporal demands of large-scale educational procurement.

Keywords:

Enrollment forecasting, fuzzy time series, forecast horizon, National Textbook Program, educational planning, learning analytics, Brazilian basic education.