Hackathon — Voice-Based Parkinson's Disease Detector

A machine learning system that detects early signs of Parkinson's Disease from voice recordings, built at a hackathon in November 2025.

How it works

1. Record a short voice sample (sustained "ahh" sound)
2. Extract 18 acoustic features (jitter, shimmer, HNR, etc.)
3. SVM classifier predicts: Healthy or Parkinson's

Pipeline

Voice Recording → Feature Extraction → SVM Classifier → Prediction

Phase	Script	Description
1	src/pd_detector.py	Load & scale UCI dataset, 80/20 split
2	src/model_trainer.py	Train SVM (RBF kernel, RandomizedSearchCV)
3	src/model_evaluator.py	Accuracy, confusion matrix, F1 score
4	src/feature_plotter.py	Visualise healthy vs PD features

Dataset

• UCI Parkinson's Dataset (195 samples, 22 voice features)
• Custom recordings from 13 participants

Tech Stack

• Python 3.11
• scikit-learn (SVM, StandardScaler, RandomizedSearchCV)
• librosa (audio feature extraction)
• pandas, numpy, matplotlib

Results

SVM with RBF kernel, 5-fold cross-validation, 50 hyperparameter iterations.

Run it

# Train the model
python3 src/pd_detector.py
python3 src/model_trainer.py

# Evaluate
python3 src/model_evaluator.py

# Predict on new voice
python3 check_person.py