Modeling Risk Factors of Acute Respiratory Infections using Logistic Regression and Multivariate Adaptive Regression Splines

Acute Respiratory Infections (ARI) remain a leading cause of morbidity among toddlers, partic ularly in regions with limited healthcare access. This study aimed to model the risk factors of ARI in toddlers using Binary Logistic Regression and Multivariate Adaptive Regression Splines (MARS). Using secondary data from Southeast Aceh, seven predictor variables were analyzed, including ma ternal characteristics, breastfeeding status, and household conditions. Both models were statisti cally significant in identifying key predictors. Logistic regression showed superior performance with 86.96% accuracy, 85.00% precision, 91.89% recall, 81.25% specificity, and 88.30% F1-score. In contrast, MARS achieved a higher recall (97.30%) but lower specificity (62.50%), indicating higher sensitivity but a greater likelihood of false positives. Exclusive breastfeeding, home ventilation, and housing density were significant predictors in both models. Overall, logistic regression was found to be the more reliable and interpretable method, offering better balance in classification metrics. These f indings support the use of logistic regression for identifying ARI risk factors in similar contexts and contribute to improved data-driven public health strategies aimed at reducing ARI incidence among vulnerable populations.