AI-Driven BMI Estimation and Personalised Health Recommendation System

Abstract

Access to basic health metrics such as Body Mass Index remains constrained for billions of people who lack proximity to clinical facilities. This paper presents an AI-driven system that estimates height, weight, and BMI from a single full-body photograph using two custom-trained U-Net neural networks operating under a multi-task learning framework on the IMDB-Wiki dataset. A pretrained Keypoint R-CNN model serves as an image quality gate before inference, generates personalised, BMI-category-specific health recommendations. The system is deployed across two production-ready platforms: a three-tier web application (React 18 + Express.js + FastAPI) and a native Android mobile application (React Native + Expo). Evaluated on 15 informal test subjects, the height model achieved a Mean Absolute Error of 6.136 cm and the weight model 9.8 kg, with 80% correct BMI category classification. All misclassifications occurred within three BMI units of a WHO category boundary. The system responds in under 3 seconds and demonstrates that meaningful preventive health screening can be delivered from a smartphone photograph alone, with no specialist equipment required