Parkinson's Disease Detection Using Acoustic features from Speech recordings

Abstract

With more than 10 million cases globally, Parkinson's disease is the second most common neurological disorder after Alzheimer's. Parkinson's disease is generally distinguished by a decline in motor and cognitive function. There isn't a single test that can be used to make a diagnosis. Instead, a thorough clinical analysis of the patient's medical history is what medical professionals must conduct. According to National Institute of Neurological Disorders research, the accuracy of an early diagnosis-defined as five years or less of symptoms-is only 53%. Although this isn't much better than winging it, prompt diagnosis is essential for successful treatment. The obtained most heavily weighted feature vector is used for classification. The proposed method makes use of the methods for linear discriminant analysis (LDA), K Nearest Neighbor (KNN), decision tree (DT), and neural network (NN). Three major contributions are made by the suggested method. Finding features from the Parkinson's acoustic dataset is the initial step. The second step is choosing the most important characteristics from the numerous produced feature vectors. The third is the addition of UPDRS-defined data, which includes motor symptoms for Parkinson's disease detection. As a consequence, KNN, LDA, DT, NN, and GB algorithms produced effective results for both acoustic characteristics and UPDRS specified data. The compared results unmistakably demonstrate that, among the chosen state-of-the-art procedures, the UPDRS established parameters attained the best success rate. Given the proposed approach and the outcomes, the approach is effective for identifying Parkinson's disease.