ML Based EEG Frequency Analysis for Alzheimer's Disease Detection

This project implements a machine learning pipeline for analyzing EEG frequency patterns to detect Alzheimer's Disease (AD). It focuses on identifying and analyzing frequency-based differences between AD patients and age-matched healthy controls using EEG data.

Data is based on research data from the article "Resting state EEG biomarkers of cognitive decline associated with Alzheimer's disease and mild cognitive impairment" which is attached to the repository.

Project Overview

The project analyzes EEG data to:

• Identify characteristic frequencies that distinguish between AD patients and controls
• Develop robust feature selection methods for EEG analysis
• Build and validate machine learning models for AD detection
• Analyze the trade-offs between different classification approaches

Repository Structure

.
├── data/                      # Data directory (not included in repo)
│   ├── processed_data.csv
│   ├── X_ml.csv
│   └── y_ml.csv
├── results/                   # Results and model outputs
│   ├── models/
│   ├── plots/
│   └── metrics/
├── exploration.py            # Initial data exploration
├── preprocessing.py          # Data preprocessing pipeline
├── feature_engineering.py    # Feature selection and engineering
├── model_training.py         # Model training and evaluation
├── report.md                 # Detailed analysis report
└── README.md                 # This file

Main Findings

The analysis identified several key frequency bands that distinguish AD patients from controls:

1. Theta (3-5 Hz) - Global
2. Delta (1-3 Hz) - Temporal regions
3. Alpha (10-13 Hz) - Central regions
4. Beta (13-20 Hz) - Parietal regions

The best performing model (CatBoost) achieved:

• 82% sensitivity
• 85% specificity
• 0.90 ROC-AUC score

Requirements

• Python 3.8+
• Required packages:
  • pandas
  • numpy
  • scikit-learn
  • xgboost
  • catboost
  • optuna
  • imbalanced-learn

Usage

1. Data Preprocessing:

python preprocessing.py

2. Feature Engineering:

python feature_engineering.py

3. Model Training:

python model_training.py

Results

Detailed results can be found in the report.md file, which includes:

• Complete feature importance analysis
• Model performance metrics
• Clinical implications
• Methodological trade-offs