A machine learning investigation of factors that contribute to predicting cognitive performance: Difficulty level, reaction time and eye-movements

Abstract

Predicting accuracy in cognitively challenging tasks has potential applications in education and industry. Task demand has been linked with increases in response time and variations in reaction time and eye-tracking metrics, however, machine learning research has not been used to predict performance on tasks with multiple levels of difficulty. We report data on adult participants who performed tasks of mental attentional capacity with six levels of difficulty and use machine learning methods to predict accuracy scores considering metrics associated with task difficulty, reaction time and eye movements. Results show that machine learning models can robustly predict performance with reaction times and difficulty level being the strongest predictors. Eye-tracking indices can also predict accuracy independently, with the most important features of the model driven by the number of fixations, number of saccades, duration of the current fixation and pupil size. Practical and theoretical implications of the results are discussed.