Seizure Prediction with Deep Learning and Machine Learning

Project Overview

This project focuses on predicting seizures using EEG data by leveraging both deep learning and machine learning models. The goal is to develop a system that can detect the onset of seizures, providing timely warnings for medical intervention. The dataset used consists of EEG recordings from pediatric subjects monitored for seizures.

Key Models Explored:

• Machine Learning Models: Support Vector Machine (SVM), Random Forest (RF), and Gradient Boosted Model (GBM).
• Deep Learning Model: Convolutional Neural Network (CNN).

Data Preprocessing

• For Machine Learning:

  • Label Encoding: Categorical labels were converted to numerical values.
  • Normalization: Features were normalized to ensure consistency across variables.
  • Data Splitting: The dataset was split into training (80%) and testing (20%) sets.

• For CNN:

  • Data Extraction: Numerical values from EEG recordings were extracted.
  • PyTorch Integration: Data was converted into PyTorch tensors to facilitate seamless model integration.
  • Normalization: Mean and standard deviation were calculated for standardizing EEG values.

🧠 CNN Model Architecture

The core of the project lies in the Convolutional Neural Network (CNN), which is particularly suited for detecting patterns in the EEG data:

• Convolutional Layers:

  • 3 convolutional blocks with kernel sizes of 16, 8, and 4, respectively.
  • Stride settings of 4, 2, and 1 for extracting fine-grained patterns from the EEG data.
  • Batch Normalization and Dropout layers after each convolution block to enhance performance and prevent overfitting.

• Fully Connected Layers:

  • Two fully connected layers that transition from convolutional operations to final decision-making, further refining the model's ability to classify seizure data.

• Activation Functions:

  • ReLU activation functions are used for non-linearity in the model.

• Optimization:

  • The model was trained using the Adam optimizer and Cross Entropy loss function, ensuring fast convergence and robust predictions.

Training & Hyperparameter Tuning

• Batch Processing: EEG data was processed in batches to optimize model training efficiency.
• GPU Acceleration: Leveraged GPU resources to speed up training.
• Hyperparameter Exploration: Different channel configurations were experimented with to maximize accuracy, resulting in an optimal configuration of 4, 2, and 4 channels for the three convolution layers.

Results

• The CNN model achieved an overall test accuracy of 74%, outperforming the other machine learning models in classifying EEG recordings for seizure prediction.
• Confusion Matrix: The confusion matrix provided insights into model performance by showing true positives, false positives, true negatives, and false negatives.
• Classification Report: Precision, recall, and F1-score metrics were computed, demonstrating the robustness of the CNN model in detecting seizures.

How to Use

1. Clone the repository:

   git clone https://github.com/YOUR_USERNAME/seizure-prediction-with-dl-ml.git