Intelligent Vitamin Deficiency Prediction Using Skin Images: A Deep Learning Approach with EfficientNet-Lite0

Chethana O S

PG Department of Computer Applications JSS College of Arts, Commerce and Science, Mysuru, India

 Mrs. Shwetha M R

Assistant Professor, PG Department of Computer Applications

JSS College of Arts, Commerce and Science,

Mysuru, India

Abstract— Vitamin deficiencies affect a large share of the global population and frequently present as visible skin, hair, and nail changes, yet confirmatory diagnosis still relies on invasive, costly, and time-consuming blood tests that are often unavailable in rural or resource-constrained settings. This paper presents an Intelligent Vitamin Deficiency Prediction system that uses deep learning and computer vision to infer likely deficiencies in vitamins A, B, C, D, and E directly from skin images. A lightweight EfficientNet-Lite0 convolutional neural network, trained with the AdamW optimizer and a CosineAnnealingLR scheduler, classifies preprocessed skin images into deficiency-related condition categories and returns a confidence score in real time. The system is delivered as a two-module web platform: an Admin module for managing users, health tips, and FAQs, and a User module through which individuals register, upload images, and receive AI-generated predictions along with preventive health guidance. Functional, integration, performance, security, and acceptance testing confirm that the system operates reliably, returns predictions within a few seconds even on standard hardware, and achieves classification accuracy comparable to prior CNN-based approaches. The results indicate that lightweight, image-based deep learning can serve as a practical, non-invasive, and low-cost screening aid that complements—rather than replaces—clinical diagnosis.

Index Terms— computer vision, convolutional neural network, deep learning, EfficientNet-Lite0, skin image analysis, vitamin deficiency prediction.