Enhancing Poshan Abhiyaan with Predictive Analytics: A Machine Learning Approach Using Anganwadi Growth Monitoring Data in Tribal Communities of Chhattisgarh, India

Abstract

Malnutrition among children remains a significant challenge in tribal regions of Chhattisgarh, India, with current reactive approaches limiting intervention effectiveness. This paper presents a comprehensive machine learning-based framework for early prediction of child nutritional status to enable timely interventions before malnutrition becomes severe. We analyzed monthly anthropometric data from Anganwadi centers in Dhamtari district using four prediction models: LSTM-FC, Random Forest, XGBoost, and Logistic Regression across a six-month forecasting horizon. Our analysis encompassed 32297 unique children with comprehensive temporal performance evaluation. Random Forest achieved the highest Month 1 accuracy of 89.7% with superior discriminative capability (ROC-AUC: 0.939), while Logistic Regression demonstrated remarkable temporal stability with only 4.4% performance decline over six months. All models achieved competitive Month 1 performance above 86%, with LSTM-FC reaching 86.6% accuracy. Critical findings revealed fundamental prediction asymmetry: models excel at monitoring existing severe malnutrition cases (F1-score: 0.915), but demonstrate limited effectiveness for incident case prediction in healthy children (F1score: 0.264). Optimal forecasting reliability occurs within three-month horizons. Our approach can transform reactive methods to proactive prevention in resource-constrained tribal settings, supporting evidence-based interventions through Poshan Abhiyaan.