Parkinson's Disease Detection from Voice Data

Overview

This project involves detecting Parkinson's disease using voice recordings. By extracting key voice features from audio data, machine learning models were trained to predict the likelihood of Parkinson's disease. The final solution includes a user-friendly Streamlit web application for real-time analysis.

---

Data Source

The dataset used for this project was created from the following source: Voice Samples for Patients with Parkinson's Disease and Healthy Controls.

Features Extracted

The audio data was processed to extract the following features for analysis:

Column Name	Description	Data Type
MDVP:Fo(Hz)	Average fundamental frequency	float64
MDVP:Fhi(Hz)	Maximum fundamental frequency	float64
MDVP:Flo(Hz)	Minimum fundamental frequency	float64
MDVP:Jitter(%)	Percentage of Jitter	float64
MDVP:Jitter(Abs)	Absolute Jitter	float64
MDVP:RAP	Relative Amplitude Perturbation	float64
MDVP:PPQ	Pitch Period Perturbation Quotient	float64
Jitter:DDP	Difference of Difference of Jitter	float64
MDVP:Shimmer	Shimmer (dB)	float64
MDVP:Shimmer(dB)	Shimmer (logarithmic scale)	float64
Shimmer:APQ3	Amplitude Perturbation Quotient (3 cycles)	float64
Shimmer:APQ5	Amplitude Perturbation Quotient (5 cycles)	float64
MDVP:APQ	Average Amplitude Perturbation	float64
Shimmer:DDA	Difference of Difference of Amplitude	float64
NHR	Noise-to-Harmonics Ratio	float64
HNR	Harmonics-to-Noise Ratio	float64
RPDE	Recurrence Period Density Entropy	float64
DFA	Detrended Fluctuation Analysis	float64
spread1	First spread measure	float64
spread2	Second spread measure	float64
D2	Nonlinear dynamical complexity	float64
PPE	Pitch Period Entropy	float64
status	Disease status (1 = Parkinson's, 0 = Healthy)	uint8

---

Machine Learning Models Trained

The following machine learning models were trained on the dataset:

1. Decision Tree Classifier
2. Random Forest Classifier
3. Logistic Regression
4. Support Vector Machine (SVM)
5. Naive Bayes
6. K-Nearest Neighbors (KNN) Classifier
7. XGBoost Classifier

Model Selection

After training and evaluating all models, the XGBoost Classifier was chosen due to its superior accuracy and ability to handle imbalanced datasets.

---

Application Development

A Streamlit app was created to make the solution accessible and interactive. Users can:

1. Upload an audio file or record their voice directly.
2. Extract features from the provided audio data.
3. Use the trained model to predict the likelihood of Parkinson's disease.

Features of the App

• Real-time audio recording and analysis.
• Display of extracted features for user insight.
• Intuitive interface for medical professionals and general users.
• Secure and efficient model prediction.

---

How to Run the App

1. Clone this repository:

   git clone https://github.com/your-repo/parkinsons-detection-app.git
   cd parkinsons-detection-app

2. Install the required dependencies:

   pip install -r requirements.txt

3. Run the Streamlit app:

   streamlit run parkinsons_app.py

---

Note

This application is for educational and demonstration purposes only. It should not replace professional medical diagnosis. Always consult a healthcare provider for medical advice.