A modified kNN algorithm to detect Parkinson’s disease

Parkinson’s disorder is the second most common neurodegenerative disease, where the patients experience unintentional agitations and motor skills deterioration over time. Therefore, the proposed work has been developed by improving the conventional kNN to detect Parkinsonian disorder with (i) Gait, (ii) Handwriting, and (iii) Voice parameters. In this approach, the concept of weights and δ-neighbourhood is introduced for predicting the class of unknown test samples by replacing voting and neighbouring points. It designs the δ-neighbourhood region under the aegis of k -neighbouring points, resulting in envisioning such a region where precise classification could be performed. To efficiently classify Parkinsonian patients from healthy individuals, the computed weights are given preference over voting at the time of testing the dataset. Hence, the suggested work has the least margin of error and is minutely sensitive towards the distribution of data for higher values of k . The average accuracy of the proposed approach is 99.60, 97.8, and 94.5% for gait, handwriting, and voice parameters, respectively . In contrast to other compared supervised classifiers, the modified kNN algorithm is more efficient in detecting Parkinson’s patients regardless of sample sizes.