Data-Driven Forecasting of Special Education Enrollment: An Explainable Machine Learning Approach

Abstract

The application of machine learning algorithms in the field of special education remains incipient despite advances achieved in other sectors. This field faces challenges related to inclusion, planning, and resource allocation, especially in contexts where administrative records are often underutilized for analytical purposes. This study proposes an explainable forecasting approach based on 23,905 historical data records to anticipate educational demand in the Special Basic Education (SBE) modality, aiming to develop and validate a Random Forest model applied to a multivariate database of official enrollment records from 2019 to 2024, projecting a slight global contraction from 28,000 to 26,800 enrollments by 2025. The findings reveal nonlinear growth patterns differentiated by region and educational level, mainly in Early SBE (ages 0 to 2), Preschool, and Primary, with a general trend of increasing demand in coastal and highland regions. The models achieved high levels of accuracy (R² > 0.97), with a Root Mean Squared Error (RMSE) below 190, a Mean Absolute Error (MAE) under 70, and a Mean Absolute Percentage Error (MAPE) below 10%. These results demonstrate the model’s utility as a strategic decision-support tool by optimizing resource planning in an education system characterized by territorial heterogeneity. The novelty of this study lies in integrating geospatial analysis and predictive algorithmic interpretability within an explainable artificial intelligence framework, fostering more equitable, transparent, and evidence-based educational planning.