Classifier performance for the decoding of decision evidence from intracranial EEG data

van Goor, Jos (2018) Classifier performance for the decoding of decision evidence from intracranial EEG data. Bachelor's Thesis, Artificial Intelligence.

Preview	Text bachelor_thesis Jos van Goor 31-03-2018.pdf Download (500kB) | Preview
	Text toestemming.pdf Restricted to Registered users only Download (81kB)

Abstract

Drift diffusion models have been successfully used to model the brain activity for decision tasks. Drift diffusion models describe the process of accumulating evidence until a certain evidence threshold is reached and a decision is made. In this exploratory research we attempted to find possible ways of classifying decision based intracranial electroencephalography (iEEG) data, and find out more about the time course of the brain's decision process. We used iEEG data from three different tasks, one visual search task, where participants needed to determine the location of a stimulus, and two match-to-sample tasks, where participants needed to compare, or recall images of faces. The participants consisted of people with implanted electrodes due to pharmalogically intractable epilepsy. Initially we tried to train a logistic regression classifier, assessing its performance with 10-fold cross-validation. This yielded results with an accuracy close to chance. In a subsequent attempt we found that feature selection improved the classifiers performance. We also found that support vector machines (SVM) were better able to predict decisions than regularized logistic regression. The SVMs performed at an accuracy consistently above 0.5, increasing towards the moment the participant responds. Finally we used the feature selection and classifier results to get a better understanding of the spatiotemporal evolution of the decision making process through the brain, and to find a pattern of increasing classifier accuracy over time to support the drift diffusion model.

Item Type:	Thesis (Bachelor's Thesis)
Supervisor name:	Vugt, M.K. van
Degree programme:	Artificial Intelligence
Thesis type:	Bachelor's Thesis
Language:	English
Date Deposited:	18 Apr 2018
Last Modified:	02 May 2018 11:44
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/16709

Actions (login required)

View Item