Data Science / AI

Childhood Excess Weight Prediction in Maranhão

The prospective half of the study: it uses the same decade of SISVAN records to forecast excess weight prevalence through 2026 across 30 series — the state, its 3 macro-regions and each of the 18 health regions. The state moves from 20.2% (2023) to 21.4% (90% CI: 20.4–22.6), but the number that matters to a health manager is a different one: 11 of the 18 health regions are rising and 6 are falling, so the statewide average hides both directions at once. Chapadinha concentrates the risk, with the highest projected prevalence (23.9%) and the largest increase (+2.2 pp). No model was picked by eye: every series ran a four-way competition (ARIMA × Prophet, each with 2020 kept or flagged as an outlier) decided purely on out-of-sample error, training through 2021 and testing against 2022–2023. The 2020 fork mattered — treating the pandemic as an outlier cut the statewide error from 4.65% to 2.52%. Submitted to II EANAPS 2026 (SES-MA / UFMA).

Tech Stack

PythonARIMAProphetTime SeriesSISVAN

Impact & Stats

30
Forecast Series
2.5%
Error (MAPE)
11 of 18
Rising Regions