Special Education Demand Prediction Database in Peru Using Machine Learning

Prediction and decision-making in various sectors have been greatly enhanced through time series analysis using machine learning algorithms. However, their application in the field of special education remains incipient. This field presents challenges in terms of inclusion, planning, and resources, especially in countries like Peru, where existing administrative records are underutilized for analytical and predictive purposes. This study explores the application of an explainable machine learning algorithm for time series modeling and forecasting in the Peruvian education system. The Random Forest algorithm was implemented on a multivariate database composed of official enrollment records in Special Basic Education corresponding to the 2019–2024 period, with the aim of projecting the evolution of demand until 2025. The demand-predictive approach allows for the identification of nonlinear and dynamic growth patterns at different educational levels (0–2 years, Preschool and Primary) across the country's regions. The models achieved high accuracy rates (R² > 0.97), with performance metrics including a root mean square error (RMSE) below 190, a mean absolute error (MAE) below 70, and a mean absolute percentage error (MAPE) below 10%, demonstrating their usefulness as a strategic support tool for decision-making, optimizing the planning of economic and financial resources for special education. Unlike traditional approaches, this proposal combines predictive modeling, algorithmic interpretability, and geospatial analysis, contributing to educational planning that is more sensitive to diversity and structural inequalities.

借助机器学习算法开展的时间序列分析，已大幅提升了诸多行业的预测与决策水平。然而此类方法在特殊教育领域的应用仍处于起步阶段。该领域在融合教育、规划与资源配置方面均存在挑战，尤以秘鲁等国家为甚——当地现有行政记录未得到充分利用，无法支撑分析与预测工作。本研究探讨了可解释机器学习算法（explainable machine learning algorithm）在秘鲁教育系统中开展时间序列建模与预测的应用。本研究针对2019至2024年的特殊基础教育（Special Basic Education）官方注册记录构建多变量数据库，并应用随机森林（Random Forest）算法，旨在预测至2025年的特殊教育需求演化趋势。该需求预测方法可识别全国各地区不同教育阶段（0-2岁、学前教育与小学教育）的非线性动态增长模式。模型实现了极高的准确率（决定系数R²>0.97），各项性能指标表现优异：均方根误差（root mean square error, RMSE）低于190、平均绝对误差（mean absolute error, MAE）低于70、平均绝对百分比误差（mean absolute percentage error, MAPE）低于10%，证明其可作为战略决策支持工具，助力优化特殊教育的经济与财政资源规划。与传统方法不同，本研究方案融合了预测建模、算法可解释性与地理空间分析，能够助力制定更具包容性、更贴合群体多样性与结构性不平等现状的教育规划方案。