Use of cognitive load measurements to design a new architecture of intelligent learning systems

In the context of learning environments, the learner’s attention and mental effort are of primary interest in the process of acquiring knowledge. Due to the skills and abilities of each learner, there is a growing need for generic and adaptive environments. In this work, we introduce a new architecture for such environments to assist the learner with a multi-agent-based approach. Using Electroencephalogram, this architecture generates learning content and interactions adapted for each learner. Moreover, this work presents a new unsupervised approach to estimate and recognize the learner’s cognitive load based on the Standardized Euclidean Distance (SED) and the Power Spectral Density (PSD) of brain rhythms within low frequencies, namely Theta[4–7 Hz] (θ) and Alpha [8–11 Hz] (α). The learner’s outcomes and estimated mental efforts are combined in the evaluation process using a fuzzy logic-based approach. Three experimental protocols are adopted in order to validate our study. These protocols are based on cognitive tasks with different difficulty levels. Experimental results show that PSD in θ and α bands in the occipital lobe accurately describe changes in the learner’s mental efforts and cognitive load according to the cognitive task difficulty level. Based on the Cohen Kappa coefficient, our cognitive load estimation approach, using α, is compared to an existing cognitive load index from the literature. This performance assessment process revealed large values (k≥0.48) in the occipital lobe, which reflects the efficiency of the proposed approach. Results from this study are mainly used in educational engineering and reeducation in order to subjectively assess the approaches and treatments offered in these contexts.