**3. Attention-aware systems**

During learning, activity maintains sustained attention important to achieve successful learning. However, it is a challenge to evaluate when students maintain their attention in learning tasks. To maintain student performance in e-learning

environments, have been developed attention-aware systems (AAS) with models that consider student's attention states. AAS systems are "capable of adapting to and supporting human attentional processes especially in situations of multi-tasking, frequent interactions with other users, and highly dynamic environments" [13].

According to D'Mello [14], the attention-aware learning technologies, in which one or more types of attention are modeled, are focused on attention. Accordantly, they should not be confused with similar systems that monitor different but related states (e.g. stress, affect, etc.). The automated attention-aware systems in e-learning settings have the advantage of estimate and respond in real-time without interrupting the learner. Typically, attention-aware intelligent systems can both access the current user focus, and make predictions concerning attention shifts. In the attention management field, the goal is on capturing the user's attentional focus, which can be built to offer personalized instruction dynamically supporting learning.

#### **3.1 Traditional sensory-based mechanisms for attention detection in e-learning**

This section is dedicated to the most recent sensory-based mechanisms concerning the attention-aware topic in e-learning. In e-learning have been used different sensory-based mechanisms for attention detection. D'Mello [15] refers to emergent technologies, in artificial intelligence in education, those related to eye-tracking and EEG devices. Eye-tracking is probably the most direct method supported by decades of scientific evidence concerning the *eye-mind link* [16] paradigm. While Brain-Computer Interfaces, such as those based on EEG, may complement or replace Eye tracking in the future. According to the same author, other indicators, such as physiology or gestures are undifferentiated signals that encode other information in addition to attention.

Typically, where someone is looking at is strongly associated with what him/she is paying attention to and think about [17]. Eye-tracking is the process of identifying where someone is looking with eye tracker equipment. Current research on multimedia learning has been used eye-tracking technology to study cognitive processes [18]. It allows to measure characteristics of eye movements; usually, there are two main types of measurements: fixations and saccades. The former reflects the attention process, while the latter reflects the change in the focus of visual attention [19]. Eye-tracking is considered one of the most direct and non-invasive ways of study attentional focus.

In a study entitled "Towards Automatic Real-Time Estimation of Observed Learner's Attention using Psychophysiological and Affective Signals: The Touch-Typing Study Case" [20] an experimental study is presented, in which attention is estimated in real-time for the touch-typing task. Results revealed that multiple linear regression models were successful to discriminate between low and high levels of attention. The proposed model is based on real-time sensory data from eye and gaze movement, pupil dilatation, and affective valences of valence and arousal. It is important to notice that this method does not take advantage of saccades and fixations typical used features of eye-tracking.

Electroencephalogram (EEG), already referred to as a "window on the mind" [21] is a physiological measurement used to examine the relationship between mental and bodily processes, in this study related to attention. EEG records the electrical activity of the brain in a non-invasive way at the scalp surface, which is a result of the summed potential currents across membranes of cells. Electrodes placed at the scalp, capture the signal most of the brain regions which are near the surface. Those signals are a) the Event-Related Potentials (ERPs) b) event-related changes in EEG activity in specific frequency bands.

In a study [22], an AAS was developed to identify low and high attention of students based on a genetic algorithm for EEG feature selection, followed by the application of the Support Vector Machines (SVM) classifier. Li et al. proposed an

#### *Improvement of Student Attention Monitoring Supported by Precision Sensing in Learning… DOI: http://dx.doi.org/10.5772/intechopen.98764*

EEG-based approach for attention recognition using k-Nearest-Neighbor Classifier (KNN) achieving an accuracy of 51.9% and 63.0% for 5-class and 3-class of attention respectively. Despite these classification rates are not high, the authors suggest use EEG along with other techniques such as pressure sensor, camera, eye tracking to have a higher accuracy rate. In a study entitled "Classification of EEG-Based Attention for Brain-Computer Interface" [23] the authors considered 4 levels of attention to be classified into different classes by an Artificial Neural Network (ANN) classifier. The accuracy, in that classification, was on average 63.5%.

Liu et al. [24] proposed a system to detect learning attention using a webcam composed of three layers: 1) image processing for face and eyes detection; 2) eyebrow region detection; 3) classifier. The system, which used SVM for classification achieved an accuracy varying between 89–93%. In a study entitled "Attention Decrease Detection Based on Video Analysis in E-Learning" [25], it is presented a scenario for analyzing individual learning attention level based on the video. It was analyzed using the OpenFace tool [26], specifically: head posture estimation, gaze focus estimation, eye movement estimation (closure and blink); mouth opening and yaw estimation; facial expression recognition. Result achieved an accuracy of 92%. Liang et al. [27] proposed a new technique to recognize human attention state using cardiac pulse from noncontact and automatic and webcam-based measurement. This approach has six different phases: 1) recording images; 2) converting images to RGB (red, green, blue) format; 3) Independent Component Analysis (ICA); 4) calculating human cardiac pulse signals using Fast Fourier Transform Algorithms (FFT); 5) featuring extraction; 6) Classification task with the algorithms: SVM, Naïve Bayes, and Gene expresser programming (GEP) based. Results revealed an accuracy of 81.82%ar in attention detection.

Artifice et al. [28], propose a methodology based on Heart Rate Variability that allows detection attention. The authors argue by hypothesis, that if we define a methodology, the authors can conduct an analysis of attention based on biosignals, then the process to determine better concentration conditions for a person can be facilitated. HRV, i.e., "the amount of heart rate fluctuations between the mean heart rate" [29], have been used to detect ECG data patterns. That variability has been studied in different target populations [30, 31]. In the field of attention, it has been proven a correlation between ECG and electroencephalogram (EEG) devices [32]. The proposed methodology for attention detection is composed of the following phases 1) pre-processing, which is dedicate to noise removal, and detection of correspondent artifacts; 2) feature extraction, refers to the extraction of HRV features, both in frequency and time domain, for further analysis; and 3) data analytics, which aims to inspect data to detect useful information that supports decision-making. A study [33], proposes an attention estimation system with modified smart glasses with inner camera for eye movement detection and, and inertial measurement for head pose position, and machine learning algorithms. Inertial measurement unit allow to acquire three-dimensional orientations, acceleration, and angular velocities. Eye tracking uses Hough transform for central point is the iris, and regions of interest allows to derive the left and right eye corners. Head pose is captured initial data from which are generated. Features, captured from eye images and perceived from IMU data are processed separately for further feature selection procedure through Sequential Floating Forward Selection (SFFS) and computed using Genetic Algorithm (GA) Support Vector Machine (SVM), in which GA optimize parameters of SVM. The system achieves an accuracy of 93.1%.

Sensory-based mechanisms for detection of user's attention in e-learning previously mentioned are synthetized concerning goals, techniques, methods, and algorithms employed, and achieved accuracy and presented in the next table (**Table 1**).



#### *Improvement of Student Attention Monitoring Supported by Precision Sensing in Learning… DOI: http://dx.doi.org/10.5772/intechopen.98764*


**Table 1.** *Sensory-based mechanisms for detection of user's attention in e-learning.*

*Computer-Mediated Communication*

*Improvement of Student Attention Monitoring Supported by Precision Sensing in Learning… DOI: http://dx.doi.org/10.5772/intechopen.98764*

Considering the current literature in the field, one can say that learner attention in e-learning environments can be estimated based on feature estimation methods acquired from devices as those previously mentioned (e.g. EEG and eye-tracker). Afterward, those features are used in machine learning models of attention enclosed in attention-aware systems.

However, such approaches do not have the appealing characteristics of newer generations of wireless network devices. The inclusion of those devices can disrupt traditional design principles, and thus revolutionize the interaction with the environment in an educational context.
