Classifier Comparison For Auditory Brainstem Responses

Expert classification of hearing levels from the Auditory Brainstem Response can be assisted in an objective manner through the use of machine learning algorithms. In this paper features are ranked and a comparison is made of the number of features used within several topical algorithms, namely, Naive Bayes (NB), KStar and Multi-Layer Perceptron. Receiver Operating Characteristic (ROC) graphs showing the relationship between sensitivity and specificity where used as a comparison mechanism. The outcome showed that reducing the input feature set from 20 to 4 did not have a significant impact on the ROC performance of all three models. Overall, NB provided the best performance, but this tailed off significantly when the input feature set was reduced to less than the top four.