09 · DEEP LEARNING · EEG · HEALTHCARE

Harmful Brain Activity Classification — EEG Seizure Detection

Overview

Built an EEG classification workflow with signal/statistical analysis, model training, fold-level validation, and final seizure versus non-seizure evaluation.

Details

Key result94% — mean classification accuracy

Links

View my notebook

Stack

PythonDeep LearningEEGSignal AnalysisCross-ValidationClassification

94%mean accuracy

93%sensitivity

95%specificity

5,508evaluated samples

Next projectVertebral Column Disease Classification →

The model reached 94% mean accuracy, 93% sensitivity, and 95% specificity with similar F1 scores for seizure and non-seizure samples.

Loss decreases while validation accuracy improves; the screenshot also shows the Fold 5 test accuracy of 0.9319.

Channel-by-channel variance shows that the EEG inputs do not all behave the same way.

Mean, standard deviation, skewness, and kurtosis were inspected across channels to understand the input signal distribution.