Online Sequential Extreme Learning Machine for Parkinson Disease Diagnosis Using Voice Data

Abstract

In recent times, it has been reported by researchers that almost 90% of people infected with (Parkinson Disease) infected people show some forms of vocal disorders when the disease is at its early stage. Consequently, the focus of the most recent researches on PD diagnosis is on the detection of vocal disorders that emerges from running speech or sustained vowel pronunciation of those who suffer from the disease. In order to be able to detect PD diseases with high accuracy, this group of researchers has employed the use of various speech processing methods to deduce clinically relevant information and feed it as input to a wide range of ML (Machine Learning) algorithms. Despite all these efforts geared towards achieving a more accurate detection rate for PD, the accuracy rates obtained by many new PD diagnosing systems remain unsatisfactory, and necessitates the need for improvement. Consequently, in this study, a PD diagnosing system is proposed using the MFCC (Mel Frequency Cepstral Coefficients) together with the OSELM (Online Sequential Extreme Learning Machine) classifier which is among the ML algorithms with high level of accuracy deployed in the classifying Parkinson’s disease. In this work, samples of voices used for experiments were acquired from the PDCD (Parkinson Disease Classification Dataset) database. The outcomes of the experimentations have revealed that the highest OSELM classifier’s performance has been accomplished with an accuracy reached up to 93.38%. Consequently, it can be concluded that the OSLEM is a technique with high efficiency for diagnosing Parkinson Disease through the use of voice data.