EEG Sleep Stages Prediction: A Hybrid LSTM and Random Forest Approach

Some previous research indicates that the stage of sleep at the moment of awakening is related to sleep inertia, as well as sleep quality and total sleep duration. These findings show that sleep inertia is typically more pronounced after waking up from non-rapid eye movement stages of sleep. This condition can frequently be accompanied by symptoms such as reduced mood and work performance, feelings of drowsiness, and insufficient decision-making. This study aims to propose a time series model combining Long Short-Term Memory (LSTM) and Random Forest (RF) to estimate the stages of sleep that will occur at this moment of the forthcoming thirty minutes. The LSTM algorithm is used to predict changes in EEG features, which helps the RF technique classify different stages of wakefulness and sleep, consisting of N1, N2, N3, and REM. This study employed two open datasets: the Sleep-EDF Database Expanded and the Haaglanden Medisch Centrum sleep staging database, to evaluate the generalizability and reliability of the method. From there, it demonstrates that the suggested framework may be widely leveraged on a variety of subjects. The average accuracy of the model for the two data sets was $93.29 \%$ and $83.48 \%$, respectively. The time computation was additionally examined. This research outcome contributes to the groundwork for applications designed for real-time prediction of sleep stages.