Implementation of Artificial Int Implementation of Artificial Intelligence in Anemia Screening for Adolescent Girls in Pontianak City: Development of a Machine Learning-Based Early Detection System

Abstract

Deficiency anemia is a major public health issue among adolescent girls in Indonesia, with a national prevalence of 32% and a higher rate in Pontianak (42.3%). Factors such as tropical climate, ethnic diversity, and limited healthcare access contribute to this condition. Conventional screening methods face challenges, including uneven laboratory availability, high costs, and low sensitivity. This study aimed to develop and evaluate an AI-based anemia screening system for adolescent girls in Pontianak, focusing on diagnostic performance, cost-effectiveness, and user acceptance. A cross-sectional, mixed-methods design was applied to 1,134 girls aged 15–18 years from 20 high schools across six districts (March–July 2025). The AI system used computer vision to analyze conjunctiva, nail bed, and facial images, combined with clinical and demographic data. Models were built using Random Forest, SVM, Neural Network, and ensemble approaches, validated against laboratory standards (hemoglobin, ferritin, transferrin saturation). Random Forest achieved 91.8% accuracy, 88.2% sensitivity, and 94.1% specificity. AI detected 52.3% more anemia cases than routine screening. Significant risk factors included low fish intake, prolonged menstruation, and underweight status. The system reduced screening costs by 79.9% and showed high user acceptance (4.2/5), proving effective and affordable for early anemia detection in Pontianak.