Performance Evaluation of Different Classification Algorithms Applied for Identifying Maternal Nutritional Status by Anthropometric Measurements

Pregnancy significantly influences infant quality and development. Maternal monitoring, indicated by body mass index (BMI) and mid-upper arm circumference (MUAC) measurements, reflects a country's socioeconomic development. Improper measurements heighten the risk of chronic energy deficiency (CED) in pregnant women and low birth weight (LBW) in infants. This study leverages artificial intelligence (AI) to enhance the detection process. Specifically, it evaluates the prediction performance of various classification methods: Decision Tree (DT), K-Nearest Neighbors (KNN), Logistic Regression (LR), Naive Bayes (NB), Random Forest (RF), and Support Vector Machine (SVM). Using interviews in Jombang District, Indonesia, these methods were expected to identify maternal nutritional status. The model design was divided into two stages: MUAC estimation generated binary classes, and BMI estimation generated multiple classes. The evaluation of these methods included various performance metrics: Accuracy (Acc), G-means, Sensitivity (Sens), Specificity (Spec), Positive Predictive Value (PPV), and Negative Predictive Value (NPV). Based on the results, all methods are proposed for both classifications, except KNN on multiple classification. KNN achieved significant scores in all matrices with p<0.01. KNN's performance is impacted by data imbalance. The study revealed a strong correlation (0.92 coefficient) between BMI and MUAC variables. The application of ML algorithms in detecting maternal nutritional status can significantly enhance the effectiveness and efficiency of health facilities, especially in areas with inadequate resources and medical personnel. However, exploring diverse ML algorithms is recommended to find optimal approaches for more varied data and to contribute solutions for sustainable development in the country.