1. Introduction
Educational resources, also known as educational economic conditions, refer to the human, material, as well as financial resources occupied, used, and consumed by the educational process [
1]. The distribution of educational resources in universities has been increasingly balanced with the progress of the social economy and the practical application of science and technology. However, achieving complete equilibrium in educational resources, especially learning environments, still takes much time and investment. Take the immersive virtual reality (IVR) device as an example, where some schools in China have built smart classrooms with virtual reality (VR) devices to help students become more immersed during learning, while others do not, which would widen the academic gap between students because of the different learning environments [
2,
3]. With the recurrence of the global COVID-19 epidemic in recent years, mobile learning has gradually emerged, but without the face-to-face guidance of teachers, students’ learning immersion experience (LIE) has become a hot issue waiting for research scholars to launch a detailed study [
4,
5].
Flow theory was proposed in 1975 by psychologist Mihaly Csikszentmihalyi. Csikszentmihalyi defines flow theory as a positive psychological state that typically occurs when people perceive a balance between the challenges associated with a situation and their ability and skills [
6,
7,
8]. Flow theory has been widely applied to evaluate human engagement in various naturalistic scenarios, such as working, gaming, painting, and learning. For example, the Privette experience questionnaire (PEQ), proposed in 1987 by Privette, was used to evaluate athletes’ immersive levels during sports activities [
9,
10], and the flow state scale (FSS), developed by Jackson and Marsh in 1996, had the same function as PEQ [
11]. A Japanese company, Hitachi, used to study whether human physiological signals would change before and after the flow experience occurred while working. The workers in the company took part in the experiment and found that their pulse rate slow down, and their breathing became more regular [
8]. This was the first time that people had begun to explore the relationship between physiological signals and immersion.
Flow happens along with nine elements: (1) challenge-skill balance, (2) action-awareness merging, (3) clear goals, (4) detailed feedback, (5) concentration on the task at hand, (6) sense of control, (7) loss of self-consciousness, (8) transformation of time, and (9) an autotelic experience [
8]. Seligman and Csikszentmihalyi stated that ‘families, schools, religious communities, and corporations, need to develop communities that foster these strengths’ [
12] (p. 8). As for educators, studying and applying the mechanisms of LIE in education and making full use of them may improve students’ learning effects and help them learn more happily. As for learners, LIE should have similar elements to a regular flow experience. In the research, we developed a detailed experimental paradigm with reference to those nine elements of flow theory, which are discussed in the next section.
Traditional assessment methods to study students’ implicit psychological states, including LIE, such as questionnaires and scales like FSS and PEQ, depend mostly on individuals’ subjective answers, making the research results lack objectivity and authenticity. Therefore, researchers start to explore how people learn with the help of the physiological recordings collected while learning, as well as advanced modern methods, and try to reveal the underlying learning mechanisms.
The fast development of biosensors and biosensing technologies has made it feasible to collect human physiological signals without pause in real-life scenarios. Recordings collected from the brain and body can be applied to indicate intentions and psychological states, which enables a physiological computing system to respond and adapt in an appropriate fashion [
13]. Nowadays, researchers can use innovative wearable biosensors, such as headbands [
14,
15,
16], wristbands [
17,
18,
19], and finger-clip detectors [
20], to record human physiological signals without interrupting what subjects are doing [
21]. Researchers have found that human cognitive ability, emotion, concentration, and engagement could be studied on the basis of different kinds of human physiological signals. These physiology-based studies are more objective and realistic than the traditional questionnaire method used to evaluate these functions [
22,
23]. Physiologists, psychologists, and educational researchers have shown increasing interest in applying wearable biosensing recordings in educational scenarios, even leading to an emerging cross-field of educational neurosciences [
24].
The state-of-the-art wearable biosensors and biosensing techniques are available for collecting data from both the central nervous system (CNS) and the autonomic nervous system (ANS) [
25]. Researchers generally used electroencephalography (EEG), or functional near-infrared spectroscopy (fNIRS) signals to characterize CNS activities. For example, EEG helps better understand the semantics of an artificial language learning task [
26]. EEG also contributes to understanding students’ mental states, interests, attention, and engagement in real classroom settings [
27,
28]. Andreas’s study found that stimulus-evoked neural responses, known to be modulated by attention, could be tracked for groups of students with synchronized EEG acquisition [
29]. Nowadays, we have non-inserted portable, low-cost headbands to record students’ signals, making it possible to develop experiments in natural learning scenarios. To represent ANS activities, heart rate (HR), pulse rate (PR), galvanic skin reaction (GSR), photoplethysmographic (PPG), skin conductance, and skin temperature are standard signals. Tools to record those signals are more convenient. For instance, in Koester and Farley’s research [
30], they observed 98 children in 3 open and 3 traditional first-grade classrooms and collected their skin conductance levels and mean PR to categorize subgroups as either high or low in physiological arousal levels. Researchers also apply those ANS signals to study students’ attention, concentration, mental state (usually stress), memory ability, and academic performance [
31,
32].
Using physiological recordings to study learning mechanisms and predict students’ academic performance holds great meaning to not only educators but also families and governments [
33,
34]. Finding how to be immersive while learning would help students learn better, leading to better academic performance and a more pleasant learning experience. According to Mihaly Csikszentmihalyi, flow is a positive psychological state and can be detected by biosensing equipment [
8]. LIE should also be a positive psychological state and might be reflected by some physiological recordings. Our research recorded students’ EEGs and PPGs and revealed the underlying secrets of students’ LIE evaluations.
With the fast development of technology, extended reality (XR) technologies, such as VR, augmented reality (AR), and mixed reality (MR), can create a good sense of immersion [
35]. IVR simulations for education have increased affective outcomes compared with traditional media [
36,
37,
38,
39,
40]. With the development of the wisdom classroom, an increasing number of VR devices are being used to improve the traditional learning environment. Our research applied VR devices to conduct a high-immersive learning environment compared with the conventional online learning environment.
Machine learning methods and deep learning methods can be both applied to analyze physiological signals. The support vector machine (SVM) belongs to machine learning methods, whose superior methods are often applied to classify physiological features. For instance, in Tan’s research [
41], they proposed a new semi-supervised algorithm combined with the Mutual-cross Imperial Competition Algorithm (MCICA), optimizing SVM for motion imagination EEG classification. Tang’s study [
42] built a measurement model for the athlete’s heart rate based on PPG recordings and SVM. As for deep learning methods, neural networks (such as convolutional neural networks, CNN, and recurrent neural networks, RNN) are often used to address human signals. In [
43], researchers empowered CNN to classify EEG signals.
In the present study, we recorded college students’ EEGs and PPGs in two learning environments: a high-immersive VR learning environment and a low-immersive online-learning environment, using a headband and a finger-clip blood oxygen probe. On the basis of SVM, we selected suitable physiological characteristics to predict LIE levels and then optimized the evaluation model.
The remainder of the paper is organized as follows: