Feature-driven machine learning to improve early diagnosis of Parkinson's disease

Abstract

Although advances in speech processing have facilitated the prognostic assessment of patients with Parkinson's Disease (PD), there is no objective method towards its early detection in a clinical setting. This study investigated the application of a novel hybrid Artificial Intelligence-based classifier for aiding early diagnosis of PD. Data on dysphonic measures and clinical scores on 68 subjects were obtained from the University of California-Irvine (UCI) Machine Learning database. Weights derived from a Multi-Layer Perceptron (MLP) were applied for feature selection and their moduli used to rank the input features based on their relative importance towards discriminating between physiological and pathological data patterns. Thus, the initial 27 features were reduced to 20 selected diagnostic factors. This reduced feature set was then input to a Lagrangian Support Vector Machine (LSVM) for classification. The overall performance of this hybrid feature-driven algorithm (MLP-LSVM) was thus compared against commercially available software and classifiers from similar studies. Results indicate an overall classification accuracy and the area under the receiver operating characteristic curve of 100% for the proposed feature-driven algorithm (MLP-LSVM), with relatively faster convergence, thus demonstrating its potential for aiding early diagnosis of PD in a clinical setting.