Journal of Education and Training Studies Vol. 3, No. 5; September 2015 ISSN 2324-805X E-ISSN 2324-8068 Published by Redfame Publishing URL: http://jets.redfame.com 242 Classification and Regression of Learner’ s Scores in Logic Environment Ramla Ghali 1 , Sébastien Ouellet 2 , Claude Frasson 1 1 Département d’Informatique et de Recherche Opérationnelle, 2920 Chemin de la Tour, H3C 3J7 Montréal, Canada 2 Department of Computer Science & Operations Research, University of Montreal, Canada Correspondence: Ramla Ghali, Département d’Informatique et de Recherche Opérationnelle, 2920 Chemin de la Tour, H3C 3J7 Montréal, Canada Received: July 9, 2015 Accepted: July 20, 2015 Online Published: August 7, 2015 doi:10.11114/jets.v3i5.1016 URL: http://dx.doi.org/10.11114/jets.v3i5.1016 Abstract This paper presents the possibility of classifying and regressing learner’s scores according to different cognitive tasks which are grouped with difficulty level, type and category. This environment is namely, Logic environment. It is mainly divided into three main categories: memory, concentration and reasoning. To classify and regress learner’s scores according to the category and the type of cognitive task acquired, we trained and tested different machine learning algorithms such as linear regression, support vector machines, random forests and gradient boosting. Primary results shows that a random forest algorithm is the most suitable model for classifying and regressing the learners’ scores in cognitive tasks, where the features most important for the model are, in descending order: the task difficulty and the task category in the case of regression, the task difficulty, the time taken by the participant before completing it, and his electroencephalogram mental metrics in the case of classification. Keywords: Cognitive tasks, task information, engagement, workload, distraction, machine-learning algorithms 1. Introduction In Intelligent Tutoring Systems (ITS) and Massive Open Online Courses (MOOC), recognition of user affective states, cognitive status and performance evolution during a task remain of great importance (Berka et al. 2004, Pope et al. 1995, Prinzel & Freeman 2000, Ramesh et al. 2014). In fact, several studies have shown that the emotional state in which a learner is placed has an impact on learning the concept (Damasio 1995, Isen 1999). This receptivity is subject to several complex parameters: emotions that are the basis of these affective states, the category of the task, the type of the task, the level of the learner, the individual differences (such as intelligence quotient) and his objectives. To detect and assess users’ alertness several studies have been undertaken in the field of artificial intelligence, human computer interaction, cognition and neuroscience (Prinzel & Freeman 2000, Wilson 2004). These works focus on using electroencephalogram (EEG) to extract more important features and bands. Three fundamental mental metrics are commonly used from EEG, namely, mental engagement, mental workload and distraction. Mental engagement is related to the level of mental vigilance and alertness during the task (high or low states of vigilance). For instance, highly challenging or difficult tasks involve more engagement. Mental workload can also be seen as the mental vigilance and cognitive load in a particular task. It was calculated according to three electroencephalogram (EEG) channels and two ratios extracted from Power Spectral Densities (Berka et al. 2004). However, distraction or drowsiness reflects the feeling of being sleepy and lethargic (Stevens et al. 2007). It was calculated mainly from Theta band. These measures intervene when a learner is involved in a task. They can reflect the degree of a learner’s concentration during a task that is necessarily depending on different types of fact ors such as a learner’s situation during the task (if he is relaxed or not), his familiarity with the presented task (the level of the learner), the type of task presented, the difficulty of the task, his motivation and emotions. All these factors lead the learner to reach a skill level that allows him or not to complete the task and acquire some knowledge. Purely calculating his score or performance on each task can assess this skill level. Despite the effort of many researchers to establish a classificati on and/or regression of a learner’s performance in some cognitive tasks (Galan & Beal 2012), it seems that it is very difficult to establish an accurate estimation of a learner’s performance due to all factors that we previously mentioned. In the same vein and in order to have an overview of a learner’s performance before accomplishing a cognitive task, we have developed Logic environment which contains seven types of cognitive tasks. These tasks are grouped into three