Design and evaluation of pediatric health prediction systems for low-resource settings

This paper presents the design and implementation of EarlyAid, a mobile-based pediatric health prediction system tailored for low-resource settings. The proposed approach combines a hybrid AI algorithm that integrates rule-based symptom mapping with a lightweight TensorFlow Lite classifier to generate condition probabilities based on caregiver- reported symptoms. The system operates entirely offline, supports multilingual input (English, Tonga, Bemba, and Nyanja), and uses age-specific visual prompts to improve usability. Unlike conventional mHealth tools that rely on cloud infrastructure and generic symptom checkers, EarlyAid is optimized for Zambian households with limited connectivity and diverse literacy levels. The novelty of this work lies in its privacy-preserving, offline-first architecture and its locally adapted symptom-to-condition mapping, which enables caregivers to receive predictive health insights without clinical supervision. Testing results show an average prediction accuracy of 89.3%, with strong caregiver feedback on cultural relevance and ease of use. EarlyAid demonstrates a scalable model for intelligent pediatric health support in underserved communities.