StellarisPat: A Novel Signum and LU-Ternary Feature Extraction Approach for Speech-Based Parkinson’s Disease Detection

Research overview

The text below is the source article’s original abstract under a verified reuse license. cc by

Abstract Parkinson’s disease classification (PDC) is a critical research area in digital health and machine learning. This study introduces a new textural feature extractor, StellarisPat, which leverages signum and ternary kernels for feature extraction. A multi-leveled feature extraction approach is employed, where levels are generated using multi level discrete wavelet transform (MDWT). StellarisPat extracts features from each level, and the most discriminative features are selected using a hybrid feature selector that combines neighborhood component analysis (NCA) and chi-square (Chi2) methods, termed NCA-Chi2. A PDC dataset was collected, comprising speech signals from Parkinson’s patients (Med On and Med Off) and healthy controls, with a total of three classes. Evaluation of the proposed technique was carried out using a two shallow classifiers: k-nearest neighbors and support vector machine. Additionally, an iterative majority voting (IMV) is applied for feature fusion to enhance classification performance. The proposed StellarisPat-based hybrid model achieved classification accuracies of 96.09% and 95.22% using k-NN and SVM, respectively, demonstrating its effectiveness for PDC.

Related topics