Detecting Vitamin A Deficiency in Schoolchildren Using an Enhanced Explainable Machine Learning Model

The most prevalent avoidable cause of vision impairments in children worldwide is vitamin A deficiency. In most cases, deficiencies can be detected through blood tests. However, blood tests are less accessible and of high cost in underdeveloped countries in Africa and Southeast Asia which hinders the efforts of detecting Vitamin A deficiency soon enough to prevent further complications.

With the development of machine learning and deep learning in risk-averse industries like healthcare and the expansion of electronic health records, there is a potential to use these techniques to arrange for a more accessible substitute to blood tests. In this study, a variety of machine learning techniques are applied to a sparse dataset of ocular symptoms and diagnoses obtained from Maradi, Niger, during routine eye exams carried out in a school environment.

The goal is to provide an affordable, accessible, and effective clinical screening system for Vitamin A deficiency in children using solely existing health records. The LGB model achieved the best accuracy: percent: 84.4, with a sensitivity of percent: 81.9, and a specificity of 84.7, outperforming prior results by almost percent: 10 in terms of accuracy, specificity, and F1-score.