Biosignal data preprocessing: a voice pathology detection application

Abstract

A methodology for biosignal data preprocessing is presented. Experiments were mainly carried out with voice signals for automa- tically detecting pathologies. The proposed methodology was structured on 3 elements: outlier detection, normality verification and distribution transformation. It improved classification performance if basic assumptions about data structure were met. This entailed a more accurate detection of voice pathologies and it reduced the computational complexity of classification algorithms. Classification performance improved by 15%.