Home / Current Issue / Paper 1719320
Modelling the Prevalence and Predictors of Childhood Obesity in Children Aged 5–10 Years Using Regularized Regression: A Cross-Sectional Study in Kogi State, Nigeria
Subject area: Science,Engineering and Technology · Area of research: Statistics
DOI: https://doi.org/10.64388/IREV9I12-1719320
Abstract
Background: Childhood obesity is rising fastest in low- and middle-income countries, yet sub-national evidence from Nigeria remains scarce. We examined the distribution and predictors of body mass index (BMI) among primary-school children aged 5–10 years in Kogi State and compared the performance of ordinary least squares (OLS) and three regularized regression estimators. Methods: In a descriptive cross-sectional design, anthropometric measurements (weight, height, BMI) and questionnaire data (age, sex, physical activity, diet type) were obtained from primary-school pupils selected by multi-stage stratified, cluster and simple random sampling. BMI was modelled with OLS, Lasso (L₁), Ridge (L₂) and Elastic Net (combined L₁–L₂) regression. Penalty parameters were tuned by cross-validation; models were compared using R², AIC, BIC, MSE and RMSE, with residual diagnostics for normality, homoscedasticity and autocorrelation. Results: Weight, height and BMI category were strong and statistically robust predictors of BMI across all four estimators (p < 0.001). Age and sex were not significant; the significance of physical activity and diet type was not supported once test statistics were recomputed from the reported coefficients and standard errors (see Note to Table 5). Elastic Net returned the most favourable fit metrics (R² = 0.987, AIC = 131.0, BIC = 150.0, MSE = 2.25, RMSE = 1.50) and the lowest variance-inflation factors among the models. All models satisfied residual assumptions (Shapiro–Wilk p > 0.05; Breusch–Pagan p > 0.05; Durbin–Watson ≈ 1.98). Conclusions: Regularized regression, particularly Elastic Net, controls multicollinearity more effectively than OLS for this anthropometric data structure. Because BMI is a deterministic function of weight and height, the very high R² should be interpreted with caution. The findings nonetheless support continued investment in school-based physical-activity and nutrition programmes, consistent with the global evidence base.
Keywords
Childhood Obesity, Body Mass Index, Elastic Net, Regularized Regression, Multicollinearity, Nigeria
How to cite this paper
@article{1719320,
author = {Sunday John Ekele, G. I. Onwuka, Tolulope O. James},
title = {Modelling the Prevalence and Predictors of Childhood Obesity in Children Aged 5–10 Years Using Regularized Regression: A Cross-Sectional Study in Kogi State, Nigeria},
journal = {Iconic Research And Engineering Journals},
year = {2026},
volume = {9},
number = {12},
pages = {3750-3756},
issn = {2456-8880},
url = {https://www.irejournals.com/formatedpaper/1719320.pdf},
abstract = {Background: Childhood obesity is rising fastest in low- and middle-income countries, yet sub-national evidence from Nigeria remains scarce. We examined the distribution and predictors of body mass index (BMI) among primary-school children aged 5–10 years in Kogi State and compared the performance of ordinary least squares (OLS) and three regularized regression estimators. Methods: In a descriptive cross-sectional design, anthropometric measurements (weight, height, BMI) and questionnaire data (age, sex, physical activity, diet type) were obtained from primary-school pupils selected by multi-stage stratified, cluster and simple random sampling. BMI was modelled with OLS, Lasso (L₁), Ridge (L₂) and Elastic Net (combined L₁–L₂) regression. Penalty parameters were tuned by cross-validation; models were compared using R², AIC, BIC, MSE and RMSE, with residual diagnostics for normality, homoscedasticity and autocorrelation. Results: Weight, height and BMI category were strong and statistically robust predictors of BMI across all four estimators (p < 0.001). Age and sex were not significant; the significance of physical activity and diet type was not supported once test statistics were recomputed from the reported coefficients and standard errors (see Note to Table 5). Elastic Net returned the most favourable fit metrics (R² = 0.987, AIC = 131.0, BIC = 150.0, MSE = 2.25, RMSE = 1.50) and the lowest variance-inflation factors among the models. All models satisfied residual assumptions (Shapiro–Wilk p > 0.05; Breusch–Pagan p > 0.05; Durbin–Watson ≈ 1.98). Conclusions: Regularized regression, particularly Elastic Net, controls multicollinearity more effectively than OLS for this anthropometric data structure. Because BMI is a deterministic function of weight and height, the very high R² should be interpreted with caution. The findings nonetheless support continued investment in school-based physical-activity and nutrition programmes, consistent with the global evidence base.},
keywords = {Childhood Obesity, Body Mass Index, Elastic Net, Regularized Regression, Multicollinearity, Nigeria},
month = {June},
doi = {https://doi.org/10.64388/IREV9I12-1719320}
}