Abstract
We compared seven forecasting models to predict student enrollment in Kazakhstan's schools using data from 2020-2024. We tested cohort component models, cohort survival models, trend regression with demographic factors, linear trend models, exponential smoothing, multi-factor regression, and weighted moving averages across 20 regions (17 regions and 3 cities) with about 3.9 million students. We measured how accurate each model was using Mean Absolute Percentage Error (MAPE). We trained models on 2020-2023 data and tested them on 2024 numbers. The linear trend model worked best, with 0.70% MAPE nationally and 0.77% MAPE across regions. Demographic models didn't work as well—cohort models did poorly at the regional level even though they have good theory behind them. Our forecasts for 2025-2027 show national enrollment growing from 3.9 million to 4.2 million students, but growth varies a lot by region. Big cities like Astana will grow 24.05% and Almaty 12.81%, while some regions will barely grow at all. Our results help educational planners pick the right forecasting methods for different areas. This study fills a research gap for post-Soviet countries where detailed forecasting evaluations are hard to find.