This research delves into the application of artificial neural networks for detecting ocular states through Electroencephalogram (EEG) signals. The primary objective was to elucidate neurophysiology and expound upon the operation of a brain-computer interface system. A Multi-Layer Perceptron (MLP) Neural Network was employed to predict ocular states, distinguishing between open and closed eyes based on EEG signals. To achieve this, a dataset comprising 2-minute EEG readings from diverse experiments, wherein participants maintained either open or closed eyes, was utilized. Subsequently, this data underwent processing, statistical tabulation, and visualization through graphing, facilitated by the Python programming language to enhance interpretability. The development of the Neural Network was conducted in Python, utilizing the Scikit-learn, TensorFlow, and NumPy libraries. The architecture of the MLP network featured 5 neurons in the input layer, along with 2 hidden layers (the first comprising 20 neurons and the second, 35 neurons). The network underwent training over 100 epochs, culminating in a model accuracy of 94.77%. This effectively validated the artificial neural network's competence in classifying ocular states. Upon scrutinizing the confusion matrix, notable instances of true positives (TP) and true negatives (TN) were observed, with 6822 cases accurately identified for closed eyes and 6825 for open eyes. Nevertheless, there were 453 instances where the model erroneously predicted closed eyes when they were open, and vice versa, transpiring on 300 occasions. In conclusion, the research affirmed the efficacy of the model through metrics such as accuracy and the analysis of the Confusion Matrix. However, it is imperative to acknowledge that this analysis does not exhaust the opportunities for applying supplementary methods to explore diverse approaches in validating the proposed model. Furthermore, it is noteworthy that comprehending the obtained results and endeavoring to formulate a more effective architecture for addressing the proposed problem can be a formidable challenge.
|
