Prediction of attention profiles at age 3 and 4 years using a machine learning approach

Abstract

Attentional development involves complex interactions between multiple cognitive processes and other systems. Individual differences in attentional tasks depend not only on age-related changes, but also on nonlinear relationships among genetic, temperament, cognitive and physical conditions, environment, and motivation. Classical statistical methods have serious constrains to address this complex nature. Therefore, this study proposes to use several machine learning algorithms (ML) to predict characteristic results of normal development and deviations in the development of executive attention at 3 and 4 years old, considering cognitive, behavioral and EEG data, and parent reported measures, collected in a longitudinal study. This approach is expected to accurately classify the attentional profiles based on the task performance, and to identify very early markers of executive attention development. This study is part of one funded longitudinal project involving an initial sample of 151 babies and their families who participated on three waves of data collection (at 6, 9, and 16-18 months-old). Two waves of data collection are added: at 36 and 48 months old. Several measures were taken, involving behavioral tasks, eye-tracking tasks, EEG/ERPs protocols, parent-reported measures of child temperament and home environment. Other measures are included in the last two waves: WPPSI-IV, spatial conflict task, sustained attention task, visual sequence learning task, delay of gratification task, EEG resting protocol, BeeAT Task, child's temperament, family SES, parenting styles, parents' mental health, and ASD/ADHD symptomatology. ML methods (e.g., artificial neural networks, fast large margin, decision trees, etc.) and time series analyses will be developed through training and cross-validation phases to study the attentional trajectories across ages. Sensitivity analyses will be carried out to provide measures of therelative importance of each predictor.