Creating a Diagnostic Tool for Parkinson's Disease Using a kNN Algorithm

We built a machine learning algorithm (kNN) to take in audio samples from a person’s voice and classify them with or without Parkinson’s. Current screening for Parkinson’s disease includes studying extensive medical history, blood tests, and neurological tests. Our goals were to develop a cost-effective screening tool for Parkinson’s using voice data rather than some more costly options like blood tests. We used the Oxford Parkinson’s Disease Detection Data Set that included 6-7 audio samples from 32 different people; 23 with Parkinsons and 9 without. This data set was analyzed for numerical data from the voices. We created an effective machine learning algorithm and determined which variables were unnecessary in the testing dataset. These goals were met with 87% accuracy, and 7 columns of unnecessary data were removed.

Funding was granted by the Adelphi University Summer Institute of Mathematical Epidemiology (ASIME).