Experimental Results for Epileptic EEG Signal Detection Models

Comprising experimental results comparing the performance of neural network models (EEGNet, AlexNet, Shallow ConvNet) combined with time-frequency transforms (CWT, STFT) for epileptic EEG signal detection. The study incorporates optimization techniques including Focal Loss, dynamic data augmentation, and an early stopping mechanism. The results show the CWT+Shallow ConvNet combination achieved optimal overall performance.

Use Cases

• Compare classification performance metrics between CWT and STFT feature extraction methods.
• Analyze the impact of the Squeeze-and-Excitation (SE) attention module on EEGNet model results.
• Evaluate the effectiveness of Focal Loss and dynamic data augmentation on model robustness metrics.
• Benchmark the precision and overall performance of CWT+Shallow ConvNet against CWT+EEGNet and other combinations.

Strengths

• Results are based on subject-independent validation, a robust evaluation method for EEG studies.
• The dataset documents comparisons of three distinct neural network architectures (EEGNet, AlexNet, Shallow ConvNet).
• Experimental design includes multiple targeted optimizations like Focal Loss and dynamic data augmentation.

Limitations

• The dataset is very small at 9.5 KB, indicating it likely contains only summary results, not raw EEG signals or full model outputs.
• Specific performance metrics, row counts, and column structures are not provided in the input.
• The data is tied to a specific experimental setup, limiting generalizability to other EEG detection tasks.

Provenance

Source

Canhui Wang via figshare.

Collection Method

Experimental results from training and validating neural network models on epileptic EEG data.

Freshness

Last updated March 20, 2026.