Deep Learning for Sleep Pattern Recognition in Wearables for Detecting Insomnia

Sleep disorders, including insomnia, represent an escalating public health issue; nevertheless, conventional diagnostic techniques such as polysomnography (PSG) are sometimes problematic for broad use. Wearable gadgets equipped with accelerometers and heart rate monitors offer a practical solution for continuous sleep monitoring. This research employs Long Short-Term Memory (LSTM), a deep learning (DL) method, to analyze wearable sensor data and identify sleep patterns to diagnose insomnia. The LSTM model was trained and validated using a dataset of sleep recordings annotated with sleep phases (wake, light, deep, REM) and insomnia classifications. The model attained a remarkable sleep stage classification accuracy of 96.75%, with precision and recall rates of 93.67% and 93.50%, respectively. The model exhibited a sensitivity of 93.67% and a specificity of 97.83% for identifying insomnia, improving upon traditional approaches by 18-20%. Moreover, the system accurately detected insomnia-related characteristics, including prolonged sleep latency and recurrent awakenings, with an error margin of approximately ±3 minutes. These findings highlight the efficacy of LSTM-based DL in delivering precise, non-invasive, and scalable sleep monitoring systems. Integrating this technology into wearable devices enables the early identification of insomnia and the delivery of personalized sleep treatments, revolutionizing sleep healthcare.