Advancing Sleep Stage Classification with EEG Signal Analysis: LSTM Optimization Using Puffer Fish Algorithm and Explainable AI

Abstract

In this study, we introduce SleepXAI, a Convolutional Neural Network-Conditional Random Field (CNN-CRF) technique for automatic multi-class sleep stage classification from polysomnography data. SleepXAI enhances classification accuracy while ensuring explainability by highlighting crucial signal segments. Leveraging Long Short-Term Memory (LSTM) networks, it effectively categorizes epileptic EEG signals. Continuous Wavelet Transform (CWT) optimizes signal quality by analyzing eigenvalue characteristics and removing noise. Eigenvalues, which are scalar values indicating the scaling effect on eigenvectors during linear transformations, are used to ensure clean and representative EEG signals. The Puffer Fish Optimization Algorithm fine-tunes LSTM parameters, achieving heightened accuracy by reducing trainable parameters. Evaluation on the Sleep-EDF-20, Sleep-EDF-78, and SHHS datasets shows promising results, with regular accuracy ranging from 85% to 89%. The proposed LSTM-PFOA algorithm demonstrates efficacy for autonomous sleep categorization network development, promising improved sleep stage classification accuracy and facilitating comprehensive health monitoring practices.