Neuro Pulse Model
This repository contains the trained model and processed dataset for predicting epileptic seizures using EEG feature data.
Model Intelligence
The model achieves exceptional performance in classifying preictal (pre-seizure) vs. interictal (normal) states:
- Accuracy: 99.12%
- F1 Score: 93.63%
- Sensitivity: 94.30%
- Specificity: 99.48%
- ROC AUC: 99.79%
The models are provided as serialized .pkl files (including master_model.pkl, ensemble_model.pkl, and xgboost_model.pkl).
Data Information
The included dataset contains fully processed and labeled EEG features derived from CHB-MIT Scalp EEG Database (patients chb01 to chb24).
- The features are provided in CSV format (
chbXX_labeled_features.csv). - Test Set Distribution: 13,356 interictal samples and 982 preictal samples.
How to Use
You can load the processed data using pandas or datasets, and load the models using joblib or pickle.
import joblib
from huggingface_hub import hf_hub_download
# Download and load the model
model_path = hf_hub_download(repo_id="JayF14/Neuro_Pulse", filename="models/trained/master_model.pkl")
model = joblib.load(model_path)