Performance Evaluation Of A Logistic Model Tree-Based System For Malnutrition Detection In Preschool Children

Abstract

Early diagnosis of malnutrition in preschool children is crucial for timely intervention and long-term health outcomes. This paper presents a detailed evaluation of a machine learning model developed using the Logistic Model Tree (LMT) algorithm for the classification of children's nutritional status into four categories: Normal, Under Nutrition, Overweight, and Micro Nutrient Deficiency. The model was trained on a dataset of 1,560 entries and optimized using attribute selection techniques. The evaluation focused on classification accuracy, confusion matrix analysis, and metrics such as precision, recall, and F1-score. The model achieved 94% accuracy, with strong macro-average values of 0.9668 (precision), 0.9319 (recall), and 0.9477 (F1-score). In addition, real-world validation was conducted on 34 pediatric cases under expert supervision, all of which aligned with model predictions. This confirms the model’s practical applicability in community-level screening for malnutrition and highlights its potential as a low-cost diagnostic tool in healthcare delivery.