Akademik Veri Yönetim Sistemi
International Conference on Contemporary Academic Research ICCAR 2023, Konya, Türkiye, 17 - 19 Mayıs 2023, cilt.1, sa.1, ss.112, (Özet Bildiri)
As a neurodegenerative disorder, Parkinson's disease affects movement and speech. We can gain
valuable insight into its progression and severity by analyzing speech patterns. In order to analyze
Parkinson's speech, a Multilayer Perceptron (MLP) is used, which is a type of artificial neural network. The
first step in analyzing Parkinson's speech with an MLP is to gather speech data from individuals with and
without Parkinson's disease. This data can include various acoustic time-frequency-based features such as
jitter(local), jitter (local, absolute), jitter (rap), jitter (ppq5), jitter (ddp). Beyond these frequency
parameters, there are also pulse parameters, amplitude parameters such as shimmer, voicing parameters,
pitch parameters and lastly harmonicity parameters such as autocorrelation and noise-to-harmonic. After
the speech data has been collected, it will need to be processed so that relevant features can be extracted
and normalized. Feature extraction techniques can include signal processing methods, statistical measures,
or machine learning algorithms in order to determine the most informative features in the speech data.
Training and testing sets are then created from the preprocessed speech data. Once the MLP is trained, it
can be used to analyze new speech samples from individuals with suspected or diagnosed Parkinson's
disease. The performance of the MLP model can be evaluated using various metrics accuracy, sensitivity,
specificity. As in the case of MLP classification results, the highest accuracy (81.05%) is obtained with
leave-one-out method by summarizing the data using the mean-standard deviation binary combination of
central tendency and dispersion metrics as in the case of MLP classification results. PD-discriminative
information has been found to be carried by sustained vows in parallel to the results reported in the
literature. These results provide an assessment of the model's ability to correctly classify Parkinson's speech
and distinguish it from non-Parkinson's speech. Assisting with Parkinson's disease diagnosis and
monitoring can be made easier using this approach.