RESUMO
In this work, a method that combines wavelet transform and Bayesian network is developed for the classification of the auditory brainstem response (ABR). First the wavelet transform is applied to extract the important features of the ABR by thresholding and matching the wavelet coefficients. A Bayesian network is then built up based on several variables obtained from these significant wavelet coefficients. In order to evaluate the performance of this approach, stratified 10-fold cross-validation is used and the network is evaluated on subject-dependent test sets (drawn from the same subjects from which the training set was derived). In particular, the data analyzed here are the ABR data with only fewer repetitions (64 or 128 repetitions) and this offers the great advantage of reducing the total time of recording, which is very beneficial to both the clinicians and the patients. Finally, a preprocessing method based on Woody averaging is applied to adjust the latency shift of the ABR data and it enhances the results.