Machine learning models for predicting malnutrition in NICU patients: A comprehensive benchmarking study

Malnutrition, affecting both adults and children globally, results from inadequate nutrient intake or loss of body mass. Traditional screening tools, reliant on detailed questionnaires, are costly, time-consuming, and often lack accuracy and generalizability. Automated machine learning (ML) alternatives promise improvements, offering efficient and versatile methods for nutritional assessment. However, they require clean, pre-processed data and careful model selection. This study aimed to evaluate the effectiveness of various ML models in predicting malnutrition. We benchmarked 22 different models for both regression and classification tasks using a malnutrition dataset. For this, a Neonatal Intensive Care Unit (NICU) patient case study was adopted. The dataset consists of 412 patients of which 232 were used to train the models. For this, a model development pipeline was developed to ensure robustness and enhance the reproducibility of the constructed models. This is achieved by testing a variety of model types including linear models, tree-based models, neural networks, and other ensemble methods. This resulted in machine learning models for both regression and classification tasks, and to optimize model efficiency by minimizing the number of required input features. The results showed that the Generalized Linear Models with Lasso or Elastic Net Regularization (GLMnet) model performed best for the regression task with an R² of 0.79, while the Extreme Gradient Boosting (XGBoost) model outperformed others in classification, achieving an Area Under the Curve (AUC) of 0.79. This study showcases alternatives to traditional screening methods to alleviate the burden on the healthcare system by providing reliable and automated methods for nutritional assessment that were compared on the same dataset.