Overview
Parkinson's disease affects motor control. It also affects the fine muscles of the larynx. The disease produces measurable changes in the frequency, amplitude, and harmonic structure of voice. This project trains nine supervised classifiers and one unsupervised method. It uses a dataset of 195 voice recordings. The project determines which approach best identifies the disease from acoustic features alone.
Problem
Parkinson's disease degrades phonation. An untrained ear cannot hear these changes. The changes are quantifiable in the signal. Detecting these changes early can support diagnosis before motor symptoms become severe. The challenge is to determine which classification approach reads these acoustic signatures most reliably. The dataset is small and high-dimensional.
Approach
This project does not train a single model and report its accuracy. Instead, it trains nine supervised algorithms and one unsupervised method. The project uses the same dataset and the same evaluation methodology for all algorithms. This produces a fair comparison. It also gives a real understanding of why some approaches work better than others for this type of problem.
The dataset consists of 195 voice recordings from 31 subjects. 23 subjects have Parkinson's disease and 8 subjects do not. Each recording has 24 acoustic features: jitter measures for cycle-to-cycle frequency variation, shimmer measures for amplitude variation, noise-to-harmonics ratios, and non-linear dynamical measures. The non-linear dynamical measures derive from chaos and fractal geometry theory.
The supervised models are Random Forest, Logistic Regression, Decision Tree, K-Nearest Neighbour, Support Vector Machine, Perceptron, Gaussian Naive Bayes, LightGBM, and XGBoost. K-Means Clustering is included as an unsupervised baseline. It reveals whether the two populations are geometrically separable without labels.
Results
The Support Vector Machine achieved the highest accuracy. The accuracy exceeds 80%. This result is consistent with the broader literature on SVM applied to the MDVP voice dataset. The kernel finds a maximum-margin separating hyperplane in a high-dimensional feature space. This ability suits the geometry that acoustic biomarkers produce.
The dataset has 195 samples. It is small enough that variance in model performance is informative about the relationship between inductive bias and data structure. Observing an SVM outperform a neural perceptron on a small, high-dimensional dataset directly is different from reading about it in a textbook.
Tech Stack
- Python