Hearing Loss Prediction in Newborns, Infants and Toddlers using Machine Learning

Hearing is one of the five senses critical to a person's day-to-day functioning. Despite enough awareness, society still has a stigma around hearing loss. It is one of the significant problems in the world today and is increasing exponentially. Early detection and intervention is the way to prevent and treat this problem. This paper focuses on predicting hearing loss in newborns, infants, and toddlers. First, the data is generated for the focused population in cooperation with an audiologist. Then, classification algorithms are applied to the data generated to build predictive models to determine hearing loss. Naïve Bayes, Support Vector Machines, XGBoost and Random Forest are the algorithms used for classification. Two datasets are generated, one with all classes having an equal number of records (balanced) and the other considering the prevalence of loss in population and noise (imbalanced). Maximum accuracy of 100% is obtained for the balanced dataset and 94.10% for the imbalanced dataset from Support Vector Machines.