Optimising Deep Learning Architectures for Insomnia Detection

Sleep is crucial for both physical and mental health. Poor sleep can lead to various disorders like insomnia, narcolepsy, and sleep movement disorders. Insomnia, characterized by difficulty falling asleep and frequent awakenings, is traditionally diagnosed using expensive and time-consuming methods like Polysomnography (PSG) and sleep diaries. Research has explored using wearable sensors like ECGs for insomnia detection, but existing methods often rely on computationally expensive techniques. This study investigates the use of raw ECG signals and standard Convolutional Neural Network (CNN) architectures and their light-weight variants for insomnia classification. The study found that a customized CNN model achieved a high F1-score of 85% for insomnia detection using raw ECG signals. This suggests that a home-based monitoring system using Internet of Medical Things (IoMT) and wearable ECG sensors could be feasible for initial insomnia screening. Individuals identified with potential insomnia should be referred to a healthcare professional for further evaluation.