Evaluating the Impact of Carboxyhemoglobin Level in the Human Body in Lagos State Using Machine Learning Algorithms

Abstract

The detection of carboxyhaemoglobin (COHb), a remarkably stable yet harmful complex present in body cells, presents a significant challenge. Elevated COHb levels can cause symptoms like headaches, nausea, dizziness, and, in severe cases, coma or death. This study utilised thirteen predictive variables, including sex, body mass index, glucose, and blood pressure. The COHb levels in Lagos State, Nigeria, were classified using various machine learning algorithms and variables. Evaluation metrics such as accuracy, precision, and confusion matrices were employed for assessment. Highly varied but negatively correlated factors significantly influenced ML predictions of COHb. Glucose was identified as the most influential predictor, due to food oxidation, it combines with oxygen and dissociates carbon monoxide from the blood. While seven out of twelve models that did not overfit during the training phase were retained, the best-performing model was an artificial neural network (ANN) with seven hidden layers of six neurons each. Apart from being the only model that correctly classified the rare individual of the fourth group by avoiding misplacement into the first group of many persons in the confusion matrix, the ANN scheme achieved the highest scores of 70% and 64% in accuracy and precision, respectively, during generalisation, alongside other optimal performances.