Título :	Implementation of artificial intelligence algorithm for EEG signals classification.
Autor(es):	Iván Alejandro García Amaya
Palabras clave:	ADHD EEG signals Machine Learning Deep Learning TDAH
Área:	INGENIERÍA Y TECNOLOGÍA
Fecha de publicación :	3-nov-2021
Facultad:	Facultad de Ingeniería
Programa académico:	Maestría en Ciencias en Inteligencia Artificial
Resumen:	Attention Deficit Hyperactivity Disorder is one of the problems that affect a considerable part of society, in this research work a couple of methodologies are proposed to address this problem through the classifications of electroencephalogram signals. The first methodology consists of filtering the electroencephalogram signals to remove the greatest amount of noise and to extract the beta band. Then it is necessary to extract the Absolute Power of the electroencephalogram signals from the selected channels (Cz, C3, C4, Fz, Pz). The Absolute Power generates a smaller data set with the help of a Genetic Algorithm which selects the most representative attributes to be able to represent them in 2D images. In this case, with the help of the Continuous Wavelet Transform technique, a Power Spectrum represented in 2D images is obtained. The three Deep Learning Models that are implemented are ResNet-152, the ResNeXt-101 and the GoogLeNet. Similar results are observed with the implementation of these models; with the ResNet-152 model an accuracy of 86.6% is shown as a result, with the ResNeXt-101 model shows an accuracy of 83.92% and with the GoogLeNet model an accuracy of 85.74 %. The second methodology for the approach to the classification of people diagnosed with Attention Deficit and Hyperactivity Disorder and people of Control making use of the electroencephalogram, proposes a filtering phase first where the noise from the signals is removed, while extracting the beta band and five different measurs are extracted: Standard Deviation, Variance, Entropy, Absolute Power and Relative Power. Each of these measurs provides relevant information. Two algorithms are implemented which are Logistic Regression and the Maximum Likelihood, these models are trained with 70% of the data set and 30% are used for validation that is randomly selected, this process is repeated several times and the best results obtained are 80.30% of accuracy in the Logistic Regression model and 84.40% in the Maximum Likelihood model, greater extraction of characteristics translates into a greater amount of information, the correct extraction of information allows us to implement algorithms such as Logistic Regression and Maximum likelihood.
URI:	http://ri-ng.uaq.mx/handle/123456789/3220
Aparece en:	Maestría en Ciencias en Inteligencia Artificial

Los ítems de DSpace están protegidos por copyright, con todos los derechos reservados, a menos que se indique lo contrario.