EEG Classification Model

This project aims to classify EEG (electroencephalogram) data into various brain activity categories, including seizure and non-seizure activities, using advanced machine learning models.

Project Objective

The primary objectives of this project are:

• To classify EEG signals into seizure and non-seizure categories (Binary Classification).
• To categorize EEG signals into five distinct classes (Multi-Class Classification):
  1. Eye-open
  2. Eye-closed
  3. Hippocampal region
  4. Epileptogenic region
  5. Seizure

Table of Contents

• Project Objective
• Dataset
• Data Preprocessing
• Feature Extraction
• Classification Models
• Evaluation and Results
• Setup and Execution
• Technologies Used
• Contributions
• License

Dataset

The Bonn EEG dataset is used for this project. The data comprises single-channel EEG recordings with 4097 features per record. Each EEG recording is classified into five distinct categories as mentioned above.

Data Preprocessing

• Metadata Creation: Efficiently created metadata to manage EEG files.
• Label Encoding & Decoding:
  • Label Encoder converts categorical labels into numerical indices for model compatibility.
  • Label Decoder facilitates interpretability by converting predictions back to original categorical labels.

Feature Extraction

The project utilizes Recurrence Quantification Analysis (RQA) to:

• Extract recurrence and determinism features from the EEG signals.
• Analyze temporal data, providing insights into repetitive and predictable patterns.
• Generate a comprehensive feature set essential for effective classification.

Classification Models

Two primary classification tasks are performed:

Binary Classification

• Task: Classify EEG signals as seizure or non-seizure.
• Approach: Simplified binary labels for seizure (0) and non-seizure (1).

Multi-Class Classification

• Task: Classify EEG signals into five activity types.
• Models Evaluated:
  • Logistic Regression
  • Random Forest
  • Decision Tree
  • XGBoost (Top performer)
• Cross-validation: 10-fold cross-validation applied to select the optimal model.

Evaluation and Results

Binary Classification

• High robustness and clear differentiation between seizure and non-seizure categories.
• Low false positives and negatives indicated by confusion matrix.

Multi-Class Classification

• Top Model: XGBoost with an F1 Score of approximately 74%.
• Class-wise performance highlights challenges, notably in capturing Classes 1 to 4 effectively.
• Overall accuracy: 26%, indicating a need for further model refinement.

Setup and Execution

Running the Project

1. Clone the repository:

   git clone <repository-url>

2. Execute provided scripts or notebooks for preprocessing, feature extraction, model training, and evaluation.

Technologies Used

• Python
• NumPy
• Pandas
• scikit-learn
• XGBoost
• Matplotlib (for visualizations)