Parkisons-Disease

Parkinson’s Disease Detection System 🧠🎙️

Welcome to the Parkinson’s Disease Detection Machine Learning Project. This tool leverages voice biomarkers—such as fundamental vocal frequencies, jitter, shimmer, and harmonic noise—to accurately predict the presence of Parkinson’s Disease using state-of-the-art machine learning algorithms.

---

🚀 Quick Start & Guides

If you’re eager to get the project running or want a deeper understanding of the code, please refer to the dedicated guide files in this repository:

1. Setup & Execution Guide (SETUP_AND_RUN.md)
   • Read this first.
   • Detailed, step-by-step instructions on Python installation, downloading module dependencies (pip install -r requirements.txt), and executing the data pipeline perfectly.
2. Technical Explanation (HOW_IT_WORKS.md)
   • Read this to understand the logic and math.
   • A deep-dive into the machine learning engineering used here: explaining data leakage prevention (Train/Test splitting), SMOTE balancing, hyperparameter fine-tuning via GridSearchCV, and Min-Max Feature Scaling.

---

📁 Repository Structure

• main.py: The core machine learning engine. Running this script downloads the remote dataset, cleans it, processes the mathematical transformations securely without data leakage, trains 7 different ML models in parallel utilizing all CPU cores, and ultimately exports the metric results into a .csv scoreboard.
• create_ppt.py: An automated Python script that reads the dynamic metrics generated by main.py and effortlessly produces a highly formatted PowerPoint (.pptx) presentation summarizing your findings.
• requirements.txt: The definitive set of Python libraries needed to run the project. Keep everything up to date by running pip install -r requirements.txt.
• Parkinsons_Project_Presentation.pptx: The automatically produced presentation output summarizing the models.
• model_metrics_comparison.csv: The spreadsheet output created by main.py, logging the performance of your Machine Learning tests.
• data.csv: A local copy of the UCI Parkinson’s dataset, automatically queried by the Python scripts when missing.
• /ML_Models/: The directory where the permanently saved/trained models (pickled .pkl records) of Random Forests, SVMs, and XGBoost are automatically dumped.

---

📊 Evaluation Overview

Our completely balanced pipeline evaluates both strict, standard baselines and powerful ensemble techniques.

Rigorous testing guarantees the data is un-leaked before modeling. As shown in the generated model_metrics_comparison.csv, our most accurate combinations currently yield:

• Support Vector Machines (SVM): ~92-94% Accuracy & F1-Score.
• XGBoost: ~89-92% Accuracy & F1-Score.
• Random Forests: ~89-92% Accuracy.

This establishes a very strong non-invasive metric for assessing early progression. Enjoy the project!