AUTHOR=Aniszewska-Stȩpień Anna , Hérault Romain , Hacques Guillaume , Seifert Ludovic , Gasso Gilles 

TITLE=Evaluating transfer prediction using machine learning for skill acquisition study under various practice conditions

JOURNAL=Frontiers in Psychology

VOLUME=Volume 13 - 2022

YEAR=2023

URL=https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2022.961435

DOI=10.3389/fpsyg.2022.961435

ISSN=1664-1078

ABSTRACT=Recent research highlighted the interest of 1) investigating the effect of variable practice on the dynamics of learning, and 2) modeling the dynamics of motor skill learning to better understand individual pathways of learners. Such modelling did not allow any prediction on future performance, both in terms of retention and transfer to new tasks. The present study attempts to quantify, by the means of machine learning algorithm, the prediction of skill transfer for three conditions of practice in a climbing task: constant practice (with no changes applied during learning), imposed variable practice (with graded contextual changes, i.e. the variants of the climbing route), and self-controlled variable practice (participants are given some control over their practice schedule of the variants). The proposed pipeline allowed us to measure adjustment of the test to the dataset, i.e. the ability of the dataset to be predictive for the skill transfer test. The behavioral data is difficult to model with statistical learning, while it is tending to be: 1) scarce (much too small data sample in comparison to the machine learning standards) and 2) flawed (data tend to contain voids in measurements). Despite these adversities, we achieved to find a machine learning pipeline working for behavioral data. The main findings show that the level of learning transfer varies, depending on the type of practice that the dynamics is pertained to: we have found that self-controlled condition is more predictive for generalization ability in learners, than the constant one.