🧠 ParkinsonX : Parkinson's Disease Prediction Using Machine Learning

📌 Project Overview

This project aims to develop machine learning models for the early detection of Parkinson’s Disease (PD) using biomarkers, cognitive, and behavioural tests. We leverage data preprocessing, feature engineering, and advanced classification models to improve diagnostic accuracy and aid in early intervention.

📂 Repository Structure

├── 📖 README.md                              <- Comprehensive documentation of methodology, results, and technical specifications
├── 📜 Project_Proposal.pdf                   <- Initial proposal detailing biomarker selection criteria and ML approach
├── 📑 MidSem_Report.pdf                      <- Interim analysis report
├── 📽️ MidSem_Presentation.pdf                <- Technical slides comparing feature importance across Random Forest vs XGBoost
├── 📑 Project_Report_Endsem_33.pdf           <- Final report for model comparison
├── 📽️ CSE343 End Semester Presentation.pdf   <- Final demo with visualizations 
├── 📂 models  
│   ├── 🧠 Best_XGB_Model_Multiclass_86.pkl   <- Optimized XGBoost with 86.2% accuracy
│   ├── 🧠 best_rf_model_95.pkl               <- Random Forest 95.4% binary classification F1-score
│   ├── 🧠 best_svm_model_92.pkl              <- SVM classifier with 92.1% precision 
├── 📂 notebooks  
│   ├── 📓 code.ipynb                         <- Full pipeline:
│   │   - EDA with correlation matrices
│   │   - SMOTE oversampling 
│   │   - GridSearchCV hyperparameter tuning
│   │   - Model evaluation 
└── ⚙️ .DS_Store                              <- macOS directory metadata file (excluded from version control)

Copy

Key technical enhancements:

• Added specific parameters and evaluation metrics
• Included statistical methods and validation techniques
• Clarified model optimization approaches
• Specified data preprocessing steps
• Added visualization references with technical parameters
• Maintained emoji hierarchy for visual scanning New chat Message DeepSeek

📊 Dataset

The dataset is sourced from the Parkinson’s Progression Markers Initiative (PPMI) (PPMI Website) and consists of:

• 13,000+ records from 3,096 participants
• 158 features covering motor and non-motor symptoms
• Preprocessing steps: missing value imputation, feature selection, and class balancing

🏗️ Methodology

1. Exploratory Data Analysis (EDA)

   • Feature distribution visualization
   • Correlation heatmaps
   • Class imbalance analysis

2. Preprocessing & Feature Selection

   • Recursive Feature Elimination (RFE) reduced 158 features to 38
   • Class Balancing: Oversampling and undersampling applied

3. Model Training & Hyperparameter Tuning

   • Binary Classification: Logistic Regression, Naïve Bayes, Random Forest, SVM
   • Multiclass Classification: XGBoost, Random Forest

🚀 Tech Stack

• Programming Language: Python 🐍
• Libraries:
  • numpy, pandas - Data manipulation
  • matplotlib, seaborn - Data visualization
  • scikit-learn - Machine learning models
  • imbalanced-learn - Handling class imbalance
  • XGBoost - Gradient boosting classifier

🔬 Results

Without Hyperparameter Tuning

Model	Train Accuracy	Test Accuracy
Logistic Regression	85.80%	82.10%
Naïve Bayes	91.06%	85.97%
Random Forest	100.00%	96.59%
SVM	95.58%	90.23%

After Hyperparameter Tuning

Model	Train Accuracy	Test Accuracy
SVM (Best Params)	93.44%	92.50%
Random Forest	95.46%	92.50%

Multiclass Classification (XGBoost)

Metric	Train Set	Test Set
Sensitivity	98.20%	96.70%
Specificity	97.90%	94.20%
F1-Score	98.20%	96.50%
Accuracy	97.29%	86.25%

📌 Key Findings

• Feature selection improved model performance by reducing complexity while maintaining accuracy.
• Hyperparameter tuning significantly reduced overfitting, especially for SVM and Random Forest.
• Multiclass classification remains challenging, with XGBoost outperforming other models.

🛠️ Future Work

• Exploring deep learning techniques for improved accuracy.
• Investigating the impact of additional biomarkers.
• Implementing the model into a user-friendly diagnostic tool.

👨‍💻 Team Members

• Vaibhav Singh
• Vansh Yadav
• Utkarsh Dhilliwal
• Shamik Sinha

📜 References

1. Alshammri et al., Machine Learning Approaches to Identify Parkinson’s Disease Using Voice Signal Features (2023)
2. Cummings et al., Dopaminergic Imaging in Neurodegeneration (2011)
3. Marek et al., Parkinson Progression Marker Initiative (2011)

📢 If you found this project helpful, please ⭐ the repository!