Obesity is a chronic, long-term medical condition that progressively worsens over time. It is a global health concern affecting both developing and developed countries. Historically, obesity was predominantly an issue among adults; however, it has increasingly become a significant problem among children and adolescents in recent years. Numerous studies have investigated the underlying risk factors influencing adolescent obesity using traditional statistical methods. However, the application of machine learning approaches remains underexplored, particularly in sub-Saharan Africa (SSA). This study aimed to apply both statistical (Logistic and probit models) and machine learning models to adolescent obesity data from Nigeria to identify key influencing variables and determine the most suitable modeling approach. The data used for this study were secondary data sourced from the Nigeria Demographic and Health Survey (NDHS) 2018. The variables analyzed included demographic characteristics and other obesity-related factors. Logistic regression, probit regression, and naïve Bayes models were employed for the analysis. The dataset consisted of a total of 491 studied participants, with 200 (40.7%) being male and 291 (59.3%) females. The summary statistics indicated that the mean age of respondents was 14.55 years, with mean systolic and diastolic blood pressures of 108.41 mmHg and 71.07 mmHg, respectively. The logistic regression analysis revealed that male adolescents were 5.727 times more likely to be obese compared to females. Additionally, the odds of being obese were 0.84 times lower for adolescents without a family history of obesity. Among the models tested, logistic regression provided the best fit for the data, demonstrating the highest sensitivity (0.8929), specificity (0.8194), area under the curve (AUC = 0.8525), and accuracy (88.80%). This model also had the lowest Akaike Information Criterion (AIC = 317.15) and Bayesian Information Criterion (BIC = 409.48). The findings of this study indicate that logistic regression is the most suitable model for identifying determinants of adolescent obesity in Nigeria. The results highlight the importance of gender and family history as significant predictors of obesity, which can inform targeted public health interventions. Further exploration of machine learning techniques in this context is recommended to enhance predictive capabilities and support evidence-based decision-making in obesity prevention efforts.
Key words: Naïve Bayes, Probit Model, Obesity, Nigeria
|
