Authors - Sanila S, S Sathyalakshmi, D. Venkata Subrahmanyan Abstract - Electroencephalography (EEG) remains the leading technique for identifying and diagnosing epileptic seizures due to its effectiveness in monitoring brain activity. However, the nonstationary nature, large volume, and rapid accumulation of EEG data present significant challenges for traditional analysis methods. To address these issues, a transition from basic data mining techniques to advanced machine learning and deep learning approaches is essential. This study focuses on developing algorithms to enhance the accuracy of seizure predictions while minimizing the volume of EEG data processed. The proposed method involves dividing EEG signals into fixed-sized windows to reduce data complexity, followed by extracting key features such as the top_k amplitude values from each window. These extracted features, combined with statistical measures of the EEG data, are then used to train classification algorithms to determine whether a seizure is occurring or not. This approach aims to balance efficiency and predictive accuracy, addressing both the computational and diagnostic challenges associated with EEG analysis. The entire raw dataset is experimented with Deep Neural Network Algorithms like Bidirectional Long Short-term memory with additional functionalities like Attention Mechanism and Spatial Weight matrix addition. Finally, both 2 D CNN and BiLSTM are applied in parallel with the additional functionalities of FFT applied EEG signal. The results are promising and found that ANN predicted with an average accuracy of 98.5%, 2DCNN with 94.3% BiLSTM with 99.6% and bi model architecture of top_k 2D CNN with Bi LSTM on FFT applied EEG signal including attention mechanism and Spatial weight matrix predicts better than all previous models with 99.8 % accuracy. .
Open Zoom
