Volume 30, Issue 2, 2021
DOI: 10.24205/03276716.2020.4055
Parkinson’s Disease Detection using Machine Learning Techniques
Abstract
Most Parkinson Disease (PD) patients suffer from vocal cord disorders. Speech impairment is an early indicator of PD. This paper focuses on development of Parkinson's disease detection system using acoustic features such as chroma STFT, RMS, spectral centroid and bandwidth, MFCC, Roll-off and Zero crossing rate which is derived from speaker's sound unit. The extracted features are modelled by various machine learning techniques. In this paper, a classification method based on Convolutional Neural Network (CNN), Artificial Neural Network (ANN) and Hidden Markov Model (HMM) are used to distinguish the PD patient’s samples from healthy people. The CNN network is trained using spectrogram of speaker information and with the acoustic features. ANN and HMM models are trained only with acoustic features. The proposed method is tested on the data obtained by the voice recordings of spontaneous dialogues and passage from participants (both PD patients and healthy controls). The recognition accuracy of CNN with spectrogram is 88% and acoustic features gives 93.5%. ANN classifier enhances the classification ability of voiceprint features such as intensity, frequency, formants and MFCC. ANN based PD achieves with the recognition rate of 96%. The HMM based Parkinson detection-based system achieves 95.2% the recognition accuracy. Hence, ANN based Parkinson detection achieves better performance compared to HMM and CNN based Parkinson detection system. The experimental results of PD based detection system prove that the proposed detection method achieves a higher accuracy of the diagnosis of PD patients from healthy people. Therefore, the above approaches can provide a solid solution for the detection of PD in the preliminary or initial stages.
Keywords
Parkinson’s Disease Detection using Machine Learning Techniques