Performance Indicators For Epileptic EEG Signal Detection Models

Comprising performance indicators from a study comparing continuous wavelet transform (CWT) and short-time Fourier transform (STFT) feature extraction methods combined with three neural network models for epileptic EEG signal detection. The study employed subject-independent validation and targeted optimizations including Focal Loss and dynamic data augmentation. The results indicate the CWT-based method with Shallow ConvNet achieved optimal overall performance.

Use Cases

• Compare the performance of CWT and STFT feature extraction methods for EEG signal classification.
• Analyze the impact of model optimizations like the Squeeze-and-Excitation module on EEGNet's classification errors.
• Evaluate the effectiveness of Focal Loss and early stopping mechanisms on model robustness across different neural network architectures.
• Benchmark the precision and overall performance of CWT+Shallow ConvNet against CWT+EEGNet and other combinations.

Strengths

• Dataset is derived from a study employing subject-independent validation, a robust methodological approach.
• Research incorporates multiple targeted optimization techniques including dynamic data augmentation and Focal Loss.
• Performance comparison involves three distinct neural network models: EEGNet, AlexNet, and Shallow ConvNet.

Limitations

• The dataset is very small at 5.5 KB, indicating it contains summary results or metrics, not raw EEG signal data.
• Specific performance metrics, row counts, and column structures for the indicators are not provided in the input.
• The dataset's scope is limited to the specific experimental setup and models described in the associated paper.

Provenance

Source

Canhui Wang via figshare.

Collection Method

Generated from experiments comparing CWT and STFT with optimized neural networks (EEGNet, AlexNet, Shallow ConvNet) for epileptic EEG detection.

Freshness

Last updated March 20, 2026.