Get Full Government Meeting Transcripts, Videos, & Alerts Forever!
Get email alerts on the School Enrollment Forecasting topic
No spam. Unsubscribe anytime.
Intern presents AI forecast that suggests future rebound for Acton-Boxborough school enrollment
Summary
Aiden McLaughlin, an intern, showed a random-forest AI model that projects enrollment in the Acton-Boxborough district will rebound from current lows and cautioned against cutting staffing or resources based only on short-term projections.
Aiden McLaughlin, an intern and rising junior at Purdue University, told the Town of Acton Finance Committee that an AI-based random forest regression model projects Acton-Boxborough school enrollment will likely rise over the next decade and could “overshoot” before stabilizing.
McLaughlin said the model used housing prices, births from five years prior, median household income, unemployment, town population and past district enrollment as input. He told the committee that the New England School Development Council (NESDC) — the district’s current forecaster — projects enrollment will fall 2.49 percent (about 123 students) by 2029 and decline 4.8 percent by 2035, but that NESDC does not publicly disclose its full formula.
The intern argued that cohort-component forecasts, which underlie many district projections, tend to lose accuracy over longer horizons because they assume linear trends in fertility, migration and other inputs. “AI allows us to dynamically input these variables to understand patterns and correlations we may otherwise be unable to predict,” McLaughlin said.
He described model performance metrics: a mean squared error equivalent to an average error of about 29 students on roughly 1,300 predictions and an R-squared of 0.99. For context, McLaughlin noted NESDC’s 2022 forecast for 2025 was off by roughly 30 students.
Committee members asked whether the model can be adapted to grade-band forecasts relevant to building decisions. McLaughlin said the model was not trained on grade-level breakdowns but “could be slightly adapted to forecast what the predictions would look like for K–6 versus high school.” He added that, if district enrollment rises, “we’d likely see a wave go through the elementary schools before we see it hit the high school.”
McLaughlin also cautioned that if the district reduced staffing, funding or other resources based solely on current projections and those projections prove too low, reversing those reductions could be difficult and time-consuming. He recommended treating AI outputs as one input among others and testing scenarios (economic downturn, boom) rather than relying on a single deterministic forecast.
The presentation prompted technical questions from committee members about software and historical backtesting; McLaughlin said he coded the model using Visual Studio Code and scikit-learn and had only recent NESDC projections (2022–2024) available for comparison.
