Predicting Unmet Contraceptive Need Among Adolescent Girls and Young Women in North-Central Nigeria: Using a Spatial Machine Learning Model
Year: 2026
Type: article
Abstract: Unmet need for contraception remains high among adolescent girls and young women in Nigeria, yet most Nigerian studies rely on conventional regression applied to older survey rounds and seldom ask where the burden falls. This study predicted unmet contraceptive need among adolescent girls and young women aged 15 to 24 years in North-Central Nigeria by combining spatial statistics with ensemble machine learning. Data came from the 2023 to 2024 Nigeria Demographic and Health Survey linked to its geographic cluster dataset, giving 1,290 young women exposed to the risk of pregnancy in 245 clusters. Spatial clustering was tested with Moran's I, local indicators of spatial association and the Getis-Ord Gi* statistic, with sensitivity analyses for coordinate displacement. A logistic regression baseline and seven tree-based and ensemble learners were compared by repeated cluster-grouped cross-validation and on a held-out test set, with outcome-based features built from training data only. Weighted unmet need was 22.1% (interval 19.3% to 25.0%), ranging from 17.4% in Niger to 35.8% in Kwara, where it was driven mainly by sexually active young women who had never married. Unmet need was spatially clustered (Moran's I = 0.0797, p = 0.013), with 15 significant hotspots, 13 of them in Kwara, a concentration that persisted across neighbourhood definitions and repeated displacement of the coordinates. The machine learning models reached a mean ROC-AUC of 0.684 to 0.713 against 0.637 for the baseline; the voting ensemble was highest, with a test ROC-AUC of 0.698, and the stacking ensemble gave the best-calibrated probabilities (calibration slope 1.03). Marital or union status and children ever born were the leading predictors. Spatial features built without outcome leakage added little to prediction (change in ROC-AUC 0.000 to +0.008). Unmet need is geographically concentrated and ensemble learning predicts it better than regression, but location adds little beyond individual and community characteristics. Programmes should prioritise the high-need area of southern Kwara, verified locally, and services for unmarried sexually active young women.
Author: Muhammad Farida Ghazali
Institution: Nasarawa State University
Language: English
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W7226011994
