Presentation Title

EEG Analysis for Predicting Early Autism Spectrum Disorder Traits

Faculty Mentor

Dr. Matin Pirouz

Start Date

23-11-2019 10:45 AM

End Date

23-11-2019 11:30 AM

Location

168

Session

poster 4

Type of Presentation

Poster

Subject Area

engineering_computer_science

Abstract

Autism Spectrum Disorder (ASD) is a neurodevelopment disorder associated with impairments in socio-communication, relationships, restrictions in thoughts, imagination etc. Autism being found as genetic in nature, depending upon person to person and their social interaction, it is important for computer science researchers to analyze the big data visualizations using phenotypic features(age, sex etc.) of each patient. The aim of this research is to develop framework in which subject does not need to push emotions. It can be done with machine learning algorithms and affective computing to produce better human-machine interface. The worldwide expeditious growth in number of ASD cases results in necessary datasets related to behavior traits. This research paper aims to deduce the emotion of disabled people through Electroencephalogram (EEG) signal by placing EEG headset electrodes on their scalp. The Machine learning algorithms help to classify the emotions and differentiate the person as autistic or neuro-typical, and extract features (wavelength, waveform, mean etc.) from EEG signals. Epochs are used to pre-train the network and fine- tune the network subject by subject. For the proposed work so far, based on datasets for Autism in toddlers and Autism in adults, a prediction model is developed which predicts probability of ASD traits so that parents/guardians can early steps. This dataset supports our hypothesis that electroencephalogram could be used to evaluate the performance of proposed methods and has potential to benefit individuals with ASD. The methods used for classification are K-nearest neighbors (KNN) algorithm, Random Forest Classifier , Support Vector Machines and Logistic regression and their performance rate is determined to choose the best classifier model with achieved precision rate of 71 percent. Experiment results show better recognition accuracy of proposed framework than traditional algorithms. These results provide the basis for the development of EEG-based brain computer interface and the proposed methods were well tolerated by different section of people- toddlers and adults.

This document is currently not available here.

Share

COinS
 
Nov 23rd, 10:45 AM Nov 23rd, 11:30 AM

EEG Analysis for Predicting Early Autism Spectrum Disorder Traits

168

Autism Spectrum Disorder (ASD) is a neurodevelopment disorder associated with impairments in socio-communication, relationships, restrictions in thoughts, imagination etc. Autism being found as genetic in nature, depending upon person to person and their social interaction, it is important for computer science researchers to analyze the big data visualizations using phenotypic features(age, sex etc.) of each patient. The aim of this research is to develop framework in which subject does not need to push emotions. It can be done with machine learning algorithms and affective computing to produce better human-machine interface. The worldwide expeditious growth in number of ASD cases results in necessary datasets related to behavior traits. This research paper aims to deduce the emotion of disabled people through Electroencephalogram (EEG) signal by placing EEG headset electrodes on their scalp. The Machine learning algorithms help to classify the emotions and differentiate the person as autistic or neuro-typical, and extract features (wavelength, waveform, mean etc.) from EEG signals. Epochs are used to pre-train the network and fine- tune the network subject by subject. For the proposed work so far, based on datasets for Autism in toddlers and Autism in adults, a prediction model is developed which predicts probability of ASD traits so that parents/guardians can early steps. This dataset supports our hypothesis that electroencephalogram could be used to evaluate the performance of proposed methods and has potential to benefit individuals with ASD. The methods used for classification are K-nearest neighbors (KNN) algorithm, Random Forest Classifier , Support Vector Machines and Logistic regression and their performance rate is determined to choose the best classifier model with achieved precision rate of 71 percent. Experiment results show better recognition accuracy of proposed framework than traditional algorithms. These results provide the basis for the development of EEG-based brain computer interface and the proposed methods were well tolerated by different section of people- toddlers and adults.