Quantum machine learning for drowsiness detection with EEG signals

Human reliability is an increasingly important area in various fields for accident prevention. Monitoring human biological parameters, such as metabolic agents, through techniques like an electroencephalogram (EEG), data analysis can help detect patterns indicating drowsiness, a major cause of fatigue that may impact tasks in various industries, including oil and gas, aviation, naval, railway, and others that involve shift work. While traditional machine learning methods, such as Multilayer Perceptron (MLP), have been explored in the literature for EEGbased drowsiness detection, advancements in computing technology have brought quantum mechanics concepts into play, offering potential advantages in computational efficiency for problem-solving. This work explores drowsiness detection via Quantum Machine Learning (QML). EEG signals are preprocessed to extract features specific to this type of data, such as Higuchi Fractal Dimension, Complexity, and Mobility, as well as statistical features such as mean, variance, root mean square, peak-to-peak, and maximum amplitude. We employ different quantum circuit architectures involving operations such as rotation gates (Ry, Rz, Ry), and entangling gates (CNOT, CZ, and iSWAP). We also combine those configurations considering different numbers of layers (1, 5, and 10). The models are trained and compared with classical MLP, considering five subjects. The main findings indicate that one subject (10) showed better results with the classic MLP model. However, for two subjects (1 and 8), iSWAP gates with 1 and 10 layers were notable, whereas for the last two subjects (5 and 6), configurations of CZ gates with 1 and 10 layers displayed the best results. This proof-of-principle study shows that QML models are suitable for analyzing EEG data related to drowsiness and can be further improved as quantum computing continues to evolve.