Clinical decision making for prediction of otitis using machine learning approach

Abstract

This study investigates the relationship between autoimmune disease otitis and gut microbial community abundance by using machine learning as an aid in the medical decision-making process. Stool samples of healthy and otitis diseased infants were obtained from the curatedMetagenomicData package. Class imbalance present in the dataset was handled by oversampling a minority class. Afterwards, we built several machine learning models (support vector machine, k-nearest neighbour, artificial neural networks, random forest and gradient boosting) to predict otitis from gut microbial samples. The best overall accuracy was obtained by the random forest classifier, 0.99, followed by support vector machine and gradient boosting algorithms, both achieving 0.96 overall accuracy. We also obtained the most informative predictors as potential microbial biomarkers for the otitis disease. The obtained results showed better accuracy in prediction of otitis from microbial metagenome than previously proposed methods found in literature.