A machine learning approach for the automatic estimation of fixation-time data signals' quality

Abstract

Fixation time measures have been widely adopted in studies with infants and young children because they can successfully tap on their meaningful nonverbal behaviors. While recording preverbal children's behavior is relatively simple, analysis of collected signals requires extensive manual preprocessing. In this paper, we investigate the possibility of using different Machine Learning (ML)-a Linear SVC, a Non-Linear SVC, and K-Neighbors-classifiers to automatically discriminate between Usable and Unusable eye fixation recordings. Results of our models show an accuracy of up to the 80%, suggesting that ML tools can help human researchers during the preprocessing and labelling phase of collected data.