Research and experiment with some activation functions in sleep stage classification using deep learning methods

Sleep disorders are becoming increasingly complex in today’ssociety. Analyzing sleep structure provides valuable information aboutsleep types, sleep cycles, and variations in different stages. However, manual sleep stage classification is challenging and time-consuming, limitingits clinical use. Deep learning methods have demonstrated promise forautomated sleep stage classification, but they face challenges such asclass imbalance and the need for interpretability. This paper investigatesthree deep learning methods, TinySleepNet, TS-TCC, and CA-TCC, using the Expanded Sleep-EDF dataset. We evaluate the performance ofthese methods and propose methods to address the aforementioned challenges. Our results show that deep learning methods can achieve highaccuracy for sleep stage classification, especially with the proposed enhancements. We also developed a sleep stage classification tool based onmethods.