Diagnosis of Parkinson's Disease Using Speech Samples and Threshold-Based Classification

摘要

In this paper we investigate the diagnosis of Parkinson's disease on the basis of characteristic features of a person's voice. First, the individual voice samples are classified as belonging either to a sick or to a healthy person. For that task, decision trees (the most efficient classifier) are selected. Second, using the threshold-based method, the final diagnosis of a person is made using previously classified voice samples. The value of the threshold determines the minimal number of individual voice samples (indicating the disease) that is required for the reliable diagnosis of a sick person. After numerous experiments with real-world data, the accuracy of classification achieved 90%. The high efficiency of diagnosis justifies that the proposed approach is worth using in medical practice.