Implementation of a quantum-based decision support system for classification of EEG signals

Electroencephalography (EEG) signals are significant biomedical data sources that measure brain activity, with applications ranging from healthcare to the analysis of daily life activities. These signals record the non-invasive electrical activity of brain waves, enabling the understanding of various neurological conditions and cognitive processes. The measurement and analysis of EEG signals involve the utilization of various machine learning and deep learning techniques. This thesis aims to analyze the efficacy of utilizing quantum machine learning techniques in the classification of EEG signals. Quantum machine learning algorithms combine the computational power of quantum computing with classical machine learning methods. The fundamental hypothesis of this thesis is that quantum machine learning algorithms can exceed classical algorithms in terms of speed and accuracy. Within the scope of this thesis, initial analyses are presented, utilizing quantum-enhanced SVM and classical machine learning algorithms on an EEG eye state dataset, thereby demonstrating the effectiveness of quantum-based classification algorithms. In a second study, the influence of various feature maps on the classification success of the QSVM algorithm is examined using an EEG schizophrenia dataset. This study achieved 100% accuracy across all qubit numbers for the Pauli X feature map. The final study within the thesis investigates VQC algorithms using different parametric quantum circuits on the LieWaves dataset. The performance of the QSVM algorithm is evaluated across state vector and Qasm simulator environments. Additionally, feature maps in QSVM and parametric quantum circuits in VQC algorithms are optimized. The results obtained are compared with the SVM algorithm, demonstrating that the proposed method outperforms classical SVM algorithms. Despite being an emerging field, the studies conducted within this thesis demonstrate that quantum machine learning algorithms have the potential to outperform classical machine learning methods. Currently, these algorithms require longer processing times compared to classical machine learning algorithms due to the utilization of quantum simulator environments. However, it is anticipated that these speed issues will be resolved with the active use of quantum computers.