Smartbands and Behavioural Interventions in the Classroom: Multimodal Learning Analytics Stress-Level Visualisations for Primary Education Teachers

ABSTRACT

Students’ stress levels may affect their well-being, attentiveness and learning outcomes in primary education classrooms. Positive behavioural interventions and support actions conducted by teachers may alleviate students’ stress levels, especially when addressing special educational needs. In this multimodal learning analytics study, students in a classroom were all given a smartband for their wrist during regular curriculum activities. Data comprised the semester of a single subject as a part of a research project conducted in Sweden. Biobehavioural stress-related arousal of students’ autonomic nervous system was visualised and analysed through distinguished behavioural modes. Additional data include naturalistic observational notes and two short teacher interviews. Research methodology and strategies for innovative implementation were presented and discussed alongside contextual details. For example, stress-level visualisations can aid actionable adjustments of behavioural intervention intensity and provide students’ attentiveness overview for teachers that sequence curricular activities during planning. Findings show an interdisciplinary basis for cost-effective real-time dynamic solutions that involve visual dashboards with advantages to understanding student learning, both at a school-wide system level and for the classroom, if viewed optimistically. However, research on the topic is still in its infancy, notably with ethical risks as a growing pain.

KEYWORDS:

• Positive behavioural interventions and support
• primary education
• wearables
• classroom
• learning analytics
• biobehavioural analysis

Introduction

A common framework for special educational change in primary education is a system-level improvement approach for learning outcomes and the well-being of students (Bryer & Beamish, 2019). Such an improvement approach includes Positive Behavioural Interventions and Support (PBIS) with School-Wide Positive Behavioural Support (SWPBS) implementations (Beamish & Bryer, 2019). Nowadays, many students have stressful life circumstances that constantly demand attention in high-stress states, increasing educational challenges as ‘their physiology also becomes persistently at odds with learning’ (Papa & Jackson, 2021, p. 125). PBIS actions conducted by teachers may be beneficial for students to regulate stress and attentiveness to maintain focus during curricular activities and may be required for many Special Educational Needs (SEN) that affect stress, attentiveness and well-being (Bryer & Beamish, 2019). However, according to Cusumano and Preston (2021), exactly when to execute PBIS actions in a primary education classroom context with student groups is notoriously unclear for teachers to identify. For example, many general labels for students with SEN may not aid teachers in identifying actionable support (Demetriou, 2022). These circumstances may be exacerbated by rapid technological innovations that sometimes come with unhealthy consequences for many students, such as smartphone addiction dysphoria (Palalas, 2018).

To emphasise the actionability of knowledge for PBIS in primary education, it is appropriate to study contextual relations of student learning through a qualitative basis for data collection methods (Algozzine et al., 2021; Potměšil & Qiong, 2022). Recent innovations in hardware capacity and sensor quality have expanded data collection sources, with increased research access to- and research interest in non-verbal behavioural data in primary education (Gunnars, 2021). This includes behaviour that previously was difficult to observe outside of clinical laboratory settings (Hogan & Baucom, 2016; Noroozi et al., 2019). For example, deployable wearables such as smartbands, smartwatches, spy glasses, and digital clothes are nowadays a feasible and affordable alternative that extends the ability of researchers in the learning sciences to collect data (Jaldemark et al., 2019). Such innovations span across education, computer science, and nursing, providing ‘interdisciplinary understandings of the human condition (e.g. biometric understanding of stress) relative to situated context’ in creating accessibility for students with SEN (Basham et al., 2020, p. 36). When emphasised as teacher assistance, such an understanding may act as a counter-argument to a reasonable prevailing fear that vital contextual details will be left out in educational applications of innovative technology (Buckingham Shum & Luckin, 2019).

Biobehavioural stress-related arousal of students’ Autonomic Nervous System (ANS) can be observed by optical smartband sensors as measured by Heart Rate (HR), electro dermal activity and recorded movement from accelerometers (Akselrod et al., 1981; Ernst, 2014; Hoog Antink et al., 2021; Noroozi et al., 2019). These readings provide unobtrusive evaluations of cognitive stress based on HR interval fluctuations, referred to as Heart Rate Variability (HRV), with the potential of providing an objective assessment of stress and resource exhaustion of the body even when not in movement or during exercise (Kim et al., 2018). For example, optical HRV monitoring through reflective Photoplethysmography (PPG) sensors have recently gained interest in scientific research as a cheap and unobtrusive method for clinical applications such as detecting cardiac arrhythmias (Hoog Antink et al., 2021). While the HRV measurements through gold standard electrocardiography are more reliable, Hoog Antink et al. (2021, p. 10) determined after rigorous comparison that PPG techniques reliably ‘could be adequate for patient monitoring’. Similarly, the monitoring of student group stress levels may be appropriate with this kind of technique. Further, Larmuseau et al. (2020, p. 1559) claim that ‘it might be useful for future research to integrate the study into regular curriculum’, as many of the current studies on HRV that relate to cognitive activity and learning collect data in experimental settings.

The biobehavioural smartband data are especially suitable for PBIS research, as descriptions based on behavioural data rather than self-reports are considered the most definitive to an applied behaviour analysis due to challenges correlating self-reports with behaviour (Kennedy, 2021; Skinner, 1950). In this case, the deployable wearables could provide the teacher with an observable indicator of student group stress and attention levels during curricular classroom sessions ‘to identify the most appropriate and effective activities’ (de Arriba-Pérez et al., 2017, pp. 326–327). Further, SWPBS depends on access to observable and measurable data for the efficacy of teachers’ PBIS actions to contextually reflect student learning and desirable learning outcomes (Cusumano & Preston, 2021). In contrast, generalised implementations often fail to accommodate student differences that is especially relevant for students with SEN (Ferri & Ashby, 2017).

Ways to analyse data have also increased and expanded with technological innovations. Researchers’ abilities to represent measurable data are expanding through new database schemas, algorithms, and intuitive user interfaces in research relating to Learning Analytics (LA) (Ifenthaler et al., 2021; Noroozi et al., 2019). Further expansion of LA by relations between various data types and attributes is emphasised in Multimodal Learning Analytics (MMLA), enabling multiple representations of data (DiMitri et al., 2018; Gunnars, 2021; Samuelsen et al., 2019). Due to the challenges of multimodal data collection, common unimodal LA approaches may solely use one data type for analysis, often collected from higher education management systems (Cukurova et al., 2020; Samuelsen et al., 2019; Viberg et al., 2018). Therefore, LA studies in primary education with multimodal applications may present a knowledge gap that seems relevant to the emphasis on actionable PBIS knowledge.

Thus, this article aims to study how biobehavioural data from deployable wearables in primary education classrooms can provide actionable knowledge for teachers conducting positive behavioural interventions relating to stress during curricular activities.

Research Question 1

What actionable knowledge for behavioural interventions can be visualised from biobehavioural data generated by deployable wearables in a primary education classroom?

Research Question 2

How can positive behavioural interventions and curricular activities in a primary education classroom relate student group stress levels with visualised biobehavioural data?

Global Increase of Behavioural Interventions

SWPBS involves multidisciplinary teams with in-depth applied behaviour analysis knowledge that aids teachers with the identification of suitable PBIS actions for students with various SEN (Beamish & Bryer, 2019; Kennedy, 2021). The interventions of similar focus have, in 2017–2022, globally increased in frequency alongside national reforms of the school curriculum, relating a higher degree of disability support previously not seen (Bryer & Beamish, 2019; Dutt et al., 2019; Hepburn, 2019; Hwang et al., 2019). While legal system reforms have garnered much attention for special education research, research on implementation and the innovative effects of such legal provisions at the school level are currently needed (Potměšil & Qiong, 2022; Tsui & Yuen, 2020).

Innovation for quality in education sectors may be fostered bottom-up through teachers during real-life curriculum activities in the classroom (Beamish & Bryer, 2019; Dutt et al., 2019; Hwang et al., 2019). However, current studies often ‘occur in settings that are not authentic or representative of real-life places where academic and social behavior are occurring’ (Algozzine et al., 2021, p. 515), which may contribute to poor translatability for actionable PBIS knowledge in top-down policy documents and related research. Knowledge actionability is crucial even to minor challenges for teachers since they are constantly limited by scarce resources, training, and materials (Algozzine et al., 2021; Beamish & Bryer, 2019; Vučinić et al., 2022).

Facilitating Special Needs in Primary Education

Student stress, well-being and performance may relate to a more thought-through version of primary education classroom management that promotes student learning flexibility. However, excessive student flexibility may be too demanding for primary education teachers by expanding the role of the teacher to a facilitator in multiple processes simultaneously (Andreozzi & Pietrocarlo, 2017). Basham et al. (2020) argue that a key driver for teacher role alterations is technological innovations, which are ongoing at a rapid pace with interdisciplinary iterations and revolutionary consequences for education. For example, smartphones may constantly put a demand on students’ attention in a way that leads to a deprioritisation of learning and other unhealthy consequences (Palalas, 2018; Papa & Jackson, 2021). Further, the teacher role now demands higher requirements of interpersonal competencies with students, parents and other teachers (Pavlidou & Alevriadou, 2022). Such conditions described above may entail connections to emotional exhaustion and teacher burnout, especially related to student groups with a high variety of SEN (Vučinić et al., 2022). Teachers and students have criticised the requirement of greater time and ability to master flexible and dynamic educational settings (Dovigo & Rocco, 2017). Further, flexible curricular activities with thought-through classroom management may require a higher awareness of contextual relations to student learning, highlighted by Algozzine et al. (2021) as lacking in PBIS research due to a common emphasis on generalisability. Such generalised approaches ‘reducing teacher decision-making and creative problem solving’ may contribute to poor recognition of individual learning differences (Ferri & Ashby, 2017, p. 23).

If certain students’ SEN are not attended to, it often leads to greater stress levels, inattention, poor engagement, and low motivation during curricular activities due to lacking well-being (Bryer & Beamish, 2019). To prevent this, PBIS actions may be conducted by teachers in inclusive classrooms to reduce student group stress levels, as followed by an analysis of student group behaviour (Cusumano & Preston, 2021). While most teachers may already perform versions of classroom management similar to applied behaviour analysis in regular curriculum activities, they may not reach the same effectiveness as a PBIS aiding SWBS team with access to comprehensive and multidisciplinary eco-psychological analysis (Hepburn, 2019). For example, such analysis may include environmental stressors such as students’ sleep, levels of happiness and frustration, and learning outcomes across multiple school subjects (Beamish & Bryer, 2019). Data on environmental stressors are commonly included in SWPBS implementations through small group instruction that involve students with SEN with PBIS aiding teams for remediation and behavioural support concerning stress reduction and student’s interpersonal functioning (Dutt et al., 2019).

One central aspect of PBIS as preventative classroom management is to relate student stress, well-being and performance by emphasising routines, clear rules for flexible student responding, and acknowledgement of appropriate behaviour through timely and context-dependent feedback in regulatory activities (Beamish & Bryer, 2019). For example, a common development for classroom management that alleviate student inattention in primary education is to combine curriculum activities with gamified elements such as competitive games on an audio-visual device (Landers, 2014). Gamified learning exemplifies how dynamic and interactive experiences with clear feedback rewards may benefit classroom management. Further, as proposed by Palmquist et al. (2021), gamification shares similarities with LA such as providing rich, algorithmically produced and context-related data.

Visualisable Multimodal Aid for Teachers

In SWPBS, students need to be correctly identified according to the three multi-level categorical tiers with different degrees of specific intervention intensity: low-intensity universal interventions in tier 1, medium-intensity targeted group interventions in tier 2, and highly intensive, individual interventions in tier 3 (Beamish & Bryer, 2019; Tutton, 2019). By supporting common behavioural challenges of tier 1, 80–90% of students may better have their social-emotional needs met at school (Beamish & Bryer, 2019). LA classroom management data could address clarity for grouping the students based on stress-related behaviour and managing their differing tiered approaches. For example, tier 2 and tier 3 consist of more frequent, intense, or targeted group-delivered interventions for the 5–10% of students that were unaffected by tier 1, often including students with SEN relating autism, deaf-blindness, intellectual disability, or social-emotional disorders (Beamish & Bryer, 2019). Thus, tier 3 requires different interventions when considering student group stress levels as compared to tier 1 and 2. Methods with LA applications could potentially provide actionable aid for narrowing down individual students’ stress through visually summarised and algorithmically adjusted data to represent students’ stress in a classroom non-invasively (Blikstein & Worsley, 2016).

Recent LA studies also produce a basis for a visual dashboard on an aiding audio-visual device for teachers that want to increase their awareness and interpretation of student-learning progress (van Leeuwen et al., 2019). The concept of a visual dashboard shows the potential for LA to provide live, context-aware data for teachers in real time during curricular activities (Cukurova et al., 2020; Ifenthaler et al., 2021). Such innovations that relate to students’ stress ‘can proactively employ technology to bridge connections between educators and students’ (Papa & Jackson,2021, p. 126). For example, the dashboard could intuitively visualise aspects of well-being and cognitive performance of student group stress- and attention levels through algorithms of smartband data relating HR and electro dermal activity analysis (de Arriba-Pérez et al., 2017; Noroozi et al., 2019). The algorithms can generate visualised smartband data that relate to foundations from clinical research.

Distinguished Stress-Related Behaviour

Foundations from clinical research differentiate short-term cardiovascular control of the ANS according to activity and recovery functions (Ernst, 2014; Hogan & Baucom, 2016). Measurement of the ANS ‘may serve as an index of stress and stress vulnerability’ (Kim et al., 2018, p. 237) by objective spectral components of rhythm and variability of HR frequency. The more cognitively demanding and performance-related sympathetic nervous system is characterised by a slow high-frequency HR, and the protective and restorative parasympathetic nervous system is characterised by a faster, lower-frequency HR (Akselrod et al., 1981; Ernst, 2014; Kim et al., 2018). As such, stress level measurements may reliably inform how behaviour with or without movement or exercise adapted to different activities alters resource exhaustion (Kim et al., 2018). While studies investigating complex learning phenomena such as specific sympathetic nervous system cognitive function demands concerning HRV analysis have had mixed results in previous research, most of them have indicated that HR and HRV generally are sensitive to stress (Johannessen et al., 2020; Larmuseau et al., 2020).

The ANS distinctions have provided a better clinical understanding of disorders that relate to resource exhaustion of the body and well-being, such as anxiety disorders that previously had been ‘dominated by models of impaired homeostasis perturbed by sympathetic overreactivity’, but better explained by a dysfunction in the parasympathetic tone (Kamath et al., 2013, p. 452). As established from special education research relating students with adverse childhood experiences, such a persistent state of high stress physiologically depletes and deprioritises the bodily resources that are required for learning (Papa & Jackson, 2021). PBIS actions, such as brief body exercises that encourage movement, may alter the high-stress cognitive state of sedentary students with depleted body resources (Luiselli, 2016).

As discussed above, smartbands with optical PPG sensors and accelerometers may present an unobtrusive alternative for providing data on the stress-related ANS (Hoog Antink et al., 2021). Similar analysis may relate PBIS actions based on readings of students’ exertion levels to evaluate appropriate PBIS intensity. According to Kennedy (2021), combining contextual details with the behavioural data type is suitable for effective PBIS actions and implementation fidelity. Thus, teachers may benefit from access to algorithmically produced and visually intuitive aiding analysis of stress-related aspects of students, similar to SWPBS, without administrative requirements. The study in this article presented below provides this kind of data analysis.

Materials and Methods

This study uses an inductive MMLA methodological approach for emphasising visualisations of actionable PBIS knowledge. As such, presented findings start bottom-up from cumulative records of biobehavioural data and relate behavioural findings only when emerging phenomena make this possible (Pierce & Cheney, 2017; Skinner, 1950). Intuitive aspects of basic statistical analysis with Statistical Package for the Social Sciences 27 (SPSS) was be heavily emphasised to promote qualitative contextual relations between the various collected data types (DiMitri et al., 2018; Potměšil & Qiong, 2022).

Xiaomi Mi Smart Band 5 with firmware version 1.0.2.66 was used as deployable wearables to determine student and teacher stress levels according to biobehavioural data in a primary education classroom in Sweden. Smartbands were rerouted and tested 6 months before data collection to functionally operate without factory-defaulted internet access requirements to meet the GDPR and national research guidelines (The Swedish Research Council, 2017). Preparation also involved locking subjects’ smartband operability and disabling device notifications during data collection. Additional security steps included database encryption during transfer and storage. Smartband management was executed with open-access software Gadgetbridge version 0.60.0. Open-access software SQLite was used as a serverless database engine.

The specific wearable device model was in September 2021 chosen as it was one of the only Gadgetbridge-compatible smartbands at the time that included a PPG sensor for optical light readings, which is a cheaper alternative for measuring HRV than the traditional electrocardiography method that requires electrodes attached with gel adhesives (Hoog Antink et al., 2021). Other beneficial factors were the affordability of the device and the toy-like form-factor suitable for primary education students, which both may be crucial if they are to be viably implemented at scale. PPG techniques should be used with caution in HRV analysis for heavy clinical analysis, as outlined by Hoog Antink et al. (2021) in the light of findings of 16% to 19% common relative errors in the low and high-frequency domain. However, for the purposes of this article that studies how deployable wearables can provide actionable knowledge through case-by-case visualisations, such relative errors could be mitigated by grouping all students through mean value estimations and confidence intervals, also seen through recommendations in the literature, e.g. (de Arriba-Pérez et al., 2017).

MMLA may encounter ethical issues and dilemmas in research relating to the extensive collection of multiple data types (Slade & Prinsloo, 2013). To adequately address national guidelines for conducting research and GDPR, very detailed steps for data collection were necessary. These steps involve relating the location and interpretation of data, informed and signed consent from all related parents, students, school leader and teacher, data deidentification, data management, and storage of data. Three separate introductory sessions without data collection comprised informed consent with the teacher, parents, and students, respectively. In acknowledging the importance of these aspects, thorough collaboration with municipalities and IT departments was maintained throughout, and the Swedish national ethical review board conducted an extensive review of detailed steps.

Data Collection

Data collection was performed in spring 2022 during a semester (5 months) of classes in a single school as a part of a larger MMLA research project conducted in Sweden. The classes contained standard national curricular contents relating to cultural knowledge, with no curricular elements of physical exercise. The school had recently performed SWPBS work based on evaluations related to the childrens’ sense of security and safety. It was currently working on raising the engagement and motivation of the students. The total number of subjects includes 24 students (around 12 years of age) and their teacher. Students were all given one smartband for their wrist during each session. Smartbands and researcher participation did not seem to affect student group behaviour beyond minimal practical concerns, according to comments provided by the teacher between sessions. Validity in contextual considerations for curricular activities was assumed to increase through constant recording conditions that solely focus on a single classroom, partly reducing complexity (Hogan & Baucom, 2016).

Every smartband recorded a cumulative record of data rows per minute with various attributes to be exported into separate database files, each session lasting around 60 minutes, generating 157 included database files in totality. After database clean-up of approximately 2200 confirmed read-error values during sessions, 10 088 relevant data rows of students were included for classroom session analysis. Eight sessions (henceforth individually referred to as ‘S#1–8’) were recorded. The first few minutes of each session before the start of curricular activities were determined as partial resting baseline measurements (Hogan & Baucom, 2016). The number of students present for each session varied between n = 19 and n = 23. Students were anonymised by being grouped in the analysis. As discussed above, the grouping also strengthened analysis in events of smartband device sensor quality issues (validity, reliability). In addition, researcher and teacher databases were collected to ensure device reliability but were not included in the analysis. See Table 1 for analysed session database rows.

Table 1. Database records and observational note word count per analysed session.

Session	S#1	S#2	S#3	S#4	S#5	S#6	S#7	S#8	TOTAL
Smartband Rows	1383	1329	1246	1094	1269	1051	1313	1403	10088
Observation Words	277	475	453	403	516	574	623	1042	4363

In conjunction, naturalistic observations, as outlined by Cohen et al. (2018), were made in each session to detail contextual relations. The contextual relations were identified by the taxonomy of multimodal data for learning as recommended by DiMitri et al. (2018), including behavioural observations, noise levels, and environment clutter, for outlining curriculum activities as brief descriptions. Each observational note-related session timestamps (for word count, see Table 1).

Two short interviews were conducted on separate occasions with the teacher after the semester was finished, netting 40 minutes of transcribed speech. Iterated statistical visualisations of smartband data were presented during the interviews to discuss actionability from the teachers’ point of view, further detailing contextual relations.

Smartband Database Coding

Each database file from the smartbands consists of five attributes, henceforth addressed as displayed in their pre-coded states: ‘TIMESTAMP’, ‘STEPS’, ‘HEART_RATE’, ‘RAW_KIND’, and ‘RAW_INTENSITY’ (Figure 1).

Figure 1. SQLite view of database workspace.

As these devices are designed to load the databases into the official app of the manufacturer, reverse engineer interpretations may not suffice to code variables. To account for this in this study, assumptions for the coding of variables are in addition to source code comparisons published on the Gadgetbridge website, based on identified patterns in database attribute functions outlined below, excluding TIMESTAMP, STEPS AND HEART_RATE which required no interpretation. However, despite thorough attempts for pattern validity, potential challenges are duly noted.

Results

Three behavioural mode variables emerged after the completed coding of RAW_KIND. Adhering to the principle of parsimony as related to PBIS actionability during curricular activities and ANS elements, RAW_KIND values were coded as ‘Movement’, ‘Focused’, and ‘Inattentive’ variables. Baseline recordings indicated that sensor readings were at least in part related HRV, as these durations had a higher prevalence of inattentive than attentive when compared to other activities. Further, comparative charts of focused and inattentive for complete sessions detailed HR patterns indicative of the ANS assumptions when separated from movement activity outlined in STEPS by the smartband accelerometer (see Figure 2 for sample).

Figure 2. Comparison of student group mean minute focused and inattentive HR values in S#1, with movement values removed. Dotted line represents regular 12-year-old HR as reference.

Inattentive shows a distinguished pattern of intensive swings between high and low HEART_RATE values and generally higher values compared to the focused behavioural mode, which adheres more closely to common HR values of subjects’ age range.

Other patterns emerged after cross-referencing these codes with RAW_INTENSITY. RAW_INTENSITY values are most likely intended to highlight exertion levels of physical exercise, as the attribute presented summarised integer values that were intuitive and visualisable in charts. Higher values denoted more intense activity. Other than relating to other attributes, the algorithm may have distinguished intensity based on HRV and electro dermal activity. While athletic workouts are outside this study’s scope, the function of RAW_INTENSITY seemed to visually differentiate stress from focused and inattentive levels better than HEART_RATE even in the absence of movement. Furthermore, it clearly indicates the strength of short PBIS movement activities.

Thus, RAW_INTENSITY and the behavioural mode variables were key components for outlining student group stress levels and providing actionable knowledge in visualised form for teachers conducting PBIS during regular curricular primary school activities.

Interpreting Intensity Levels for Actionable Knowledge

To relate student group stress-level intensity charts, the mean value of RAW_INTENSITY with a 95% confidence interval was visually represented according to each related minute, combined with markings relating the start of curricular activities for each separate session. The recordings started before the teacher-led activities, included in the charts partly as a contextual baseline. Charts end when the teacher signals for a break, cancelling the final curricular activity, and students leave the classroom. See Figure 3 for S#4.

Figure 3. Mean minute value of student group stress from S#4 with 95% CI. Box 1 denote start of the first curricular activity, ending with the start of the second curricular activity in box 2, and so on.

Most sessions comprised four curricular activities. To relate PBIS elements, a teacher can evaluate student group stress levels depending on curricular activity length and sequence. For example, fluctuating intensity durations might indicate the previous activity as too demanding, which likely would be a good time for focused behavioural mode reinforcement through a PBIS movement activity.

One pattern that directly stands out for the viewer of the visualisations is that curricular activity shifts often influence the student group to increase stress levels, only to drop moments again after. This may indicate that activity shifts, in most cases, alter stress regulation as spikes. For example, in S#4 (Figure 3), the introduction of three of the four activities seemed to follow this pattern. The first two activities comprised individual work, where quiet focus was desirable. At the end of the second activity, some students presented high-stress levels, indicating that the focus requirements of the activity were too high for some students. In this situation, the teacher that considers student group stress levels could decide to start a PBIS movement exercise to reinforce lower stress levels of the students in need of movement or decide to expose the student group to the challenge of prolonged focus activity.

Interpreting Behavioural Mode Session Totals for Actionable Knowledge

To relate student group stress-related behavioural modes, session total values were calculated, outlined in Table 2.

Table 2. Student movement, inattentive and focused behavioural mode session totals.

Session:	S#1	S#2	S#3	S#4	S#5	S#6	S#7	S#8
Focused	77,4%	78,9%	68%	80,3%	74,6%	74,9%	78,5%	68,9%
Movement	4,2%	3,5%	10,3%	2%	1,3%	2,9%	2,7%	10,1%
Inattentive	18,4%	17,7%	21,7%	17,6%	24,1%	22,2%	18,8%	21%

In conjunction with the produced from data detailing session behavioural mode totals, a teacher could compare sessions and take action to adjust the activities for desired symmetry according to the total stress level of the student group per session.

For example, during S#5, the movement value (1.3%) was lower than the rest of the sessions, thus potentially in need of a PBIS activity that would reinforce movement frequency. In contrast, S#3 and S#8 showed high movement (21,7%, 21%) yet performed at about the same level of focus as S#5 (68%, 68,9% as compared to 74,6%), indicating diminishing stress-reducing returns from the PBIS movement activities.

Relating Curricular Activities to Student Group Stress Levels and Stress Related Behaviour

According to observed phenomena in observational notes, curricular activities related to stress, well-being, behavioural mode variables, and PBIS interventions could be categorised according to social contextual details, focus expectations, and learning outcome assessment requirements.

Generally, activities with expectations of learning outcomes mainly comprised (a) individual assignments or (b) verbal contributions in active group activities. Other activities were either (c) passive group activities, such as watching a movie in front of the class relating to the curricular content or (d) behavioural interventions with or without relation to curricular content, such as brief movement activities for students with ADHD, playful teambuilding activities of throwing a softball to each other, or a game of ‘Kahoot!’ as closing activity of the session. These activity types will henceforth be referred to as ‘(a–d)’ (Table 3).

Table 3. Typology of curricular activities as emerged from phenomena in observational notes.

Individual assignments with expectations of learning outcomes. Active group activities with expectations of verbal contributions as learning outcome. Passive group activities with no expectations of learning outcomes. Behavioural interventions with or without relation to curricular content.

Kahoot! is an example of gamified (d) where students compete against each other by a mediating audio-visual device in front of the class. While situated and presented as a (d), interviews revealed that teacher also, in smaller part, view Kahoot! as an (a).

Based solely on the visualised stress levels, a teacher could evaluate the student group stress levels effects of the behavioural interventions and the sequential structure of various sessions according to curricular activity types and length. For example, S#4 (Figure 3) comprised the sequence (a), (a), (c), (b), which might indicate that such a sequence pattern of session structure is appropriate, as S#4 was the least intensive session and did not seem to require any (d).

Another example relating to the curricular activities and previously discussed S#8 showed high movement (see Figure 4).

Figure 4. Mean minute value of student group stress from S#8 with 95% CI. Boxes denote the start of curricular activities: 1(d) ‘change furniture arrangements’, 2(a) ‘study textbooks’, 3(a) ‘make short summaries of textbook content part 1/2’, 4(d) ‘run around school building’, 5(a) ‘make short summaries … part 2/2’, 6(b) ‘discussion of content’.

Looking at the chart of S#8, a movement activity was clearly indicated as very intense by the large intensity spike compared to similar PBIS activities (see Figure 4 ‘4(d): Run around school building’).

In contrast, S#5 might be considered the most stressful session, as it had high inattentive levels (17,6%) and low movement (1,3%). S#5 comprised many group activities, with the overall structure (c), (b), (b), (c), (b). This session also differs more in the highest and lowest intensity values between the mean, with a bias towards higher values, indicating that fewer students preferred group activities than individual ones, at least when placed in the specified sequence pattern of the session structure. For a chart of S#5, see Figure 5.

Figure 5. Mean minute value of student group stress from S#5 with 95% CI. Boxes denote the start of curricular activities: 1(c) ‘planning of future activities’, 2(b) ‘read aloud from textbook, 3(b) ‘discussion of content’, 4(c) ‘watch movie’, 5(b) ‘discussion of class structure’.

Another finding available through visual examination is the similarities across S#1, S#2, S#6, and S#7. In these sessions, the students were told at the beginning that they might get to play Kahoot!. In all these instances, the children performed their assigned activities in a timely manner and closed the session structure with Kahoot!. Additionally, three out of four of these sessions presented strong session total focused values (78,9%, 78,5%, 77,4%).

Teacher Beliefs of Visualisations

Visualisations like Figures 3–5 of all sessions were presented for the teacher interview occasions, along with student behavioural mode variables movement, focused, and inattentive codes visually represented as pie charts from session totals values (see Table 2). During these interviews, intervention actionability based on the student group stress levels were discussed.

The initial reaction to the charts and pie-charts was overwhelmingly positive: ‘when I see these graphs, I get very proud of my students, they seem to have greater [abilities to hold extended] focus than I expected [of them]’.

The teacher highlighted the S#4 slide (Figure 3) and corresponding pie chart as the first mentioned example of particular interest where students performed better than expected: ‘you know, this is very interesting, because after this session, the kids complained how stressful and anxiety-inducing it was, but this shows that they actually were focused despite believing that they were not’.

During S#4, the teacher used tablets with screen recordings in conjunction with an exam: what is interesting about this session is that the exam were fully done on their own iPads, and if I interpret this graph correctly, it seemed to work just fine. Previously this was a group assignment, which required a lot of verbal [−specific SEN] support.

Another aspect is the social relations of the student group. Teacher comments on inattentive levels in S#5 (Figure 5 and Table 2): ‘these kids that appeared to fight [during 3(b) “discussion of content”] are actually close friends, so they dare to maybe rile each other up in a way that they do not do to other peers in the group’.

The initial positive reactions had some exceptions, however, such as two slides that did not spark any interest: ‘well, this slide did not tell me anything I did not know, really’. Thus, conducting these readings may not always be worth the effort.

Summary of Findings

The findings of this study may be summarised accordingly:

• Biobehavioural data collected from students in primary education classrooms with deployable wearables appeared to provide objective readings of the ANS and body exertion stress-level intensity.

• By intuitively visualising curricular activities, the objective readings provided actionable knowledge for the evaluation of positive behavioural interventions, such as determining appropriate movement intensity and estimated attentiveness challenge according to activity duration and type.

• In-depth analysis of the objective readings alongside observational notes provided plentiful findings that related contextual aspects, such as consistent intensity spikes from activity shifts and optimistic indications of gamified Kahoot! elements.

Conclusion and Discussion

To conclude, this study has, alongside contextual classroom details, analysed MMLA visualisations of the ANS from related biobehavioural data to outline the body exertion intensity of students’ stress levels during regular curricular activities and promote actionable PBIS knowledge. Further, the biobehavioural data collected from primary education classrooms with deployable wearables appeared to distinguish student behavioural modes that require cognitive performance for longer durations. These behavioural modes include focused, inattentive, and movement distinctions, attributable to each minute for session total per cent. These key components related to student group stress levels that, when visualised, showed potential for actionable PBIS knowledge. For example, after a quick glance at visualisations, the intensity of curricular activities with PBIS elements can be adjusted and related in a behaviour analysis that considers students’ stress levels. Additionally, the visualisations were met with overwhelmingly positive teacher reactions, with the immediate realisation of student group stress evaluation regarding their ability to focus during extended durations.

To further conclude, in-depth analysis alongside observational notes revealed curricular activity types suitable for sequencing of activities according to stress levels. One apparent and notable finding across all visualisations indicates that activity shifts in most cases alter stress regulation, with observable and consistent intensity spikes. This finding of intensity spikes between curricular activities mirrors research on students with SEN relating executive functions that highlight the challenges of shifting attention to another cognitive state or task (Bishara & Kaplan, 2022). Other findings include prolonged activities that were challenging for the students’ ability to regulate their stress intensity levels. Thus, both curricular activity type and length seemed vital when considering session activity sequences.

Specific contextual findings of curricular activities related to student group stress levels were observed, including optimistic indications of curricular activities with gamified elements such as Kahoot!, a preference of the student group towards individual activities rather than group activities, an example of a PBIS movement activity that perhaps was too strong, and an occasion where social dynamics of a few students influenced peer students’ stress levels.

Assessment of sensor quality may require further studies but did not seem to pose an issue for the aims of this article when the analysis considers the student group as a whole, in addition to thorough database clean-up.

Methodological Implications for Research and Practice

The multimodal data analysis of this study, and similar types, are ripe with uncharted opportunities, with obvious advantages to a greater understanding of student learning (Ifenthaler et al., 2021), both at the school-wide system level and for curricular activities of a single subject. For example, the pie-chart visualisations of the three identified behavioural mode variables was in this study analysed at the level of curricular activity length and sequences of the data collection classroom site. However, the charts could also emphasise SWPBS work when combined with other recordable contextual details. For example, light levels, noise, heat, humidity, and air pollution may be relevant for such a multimodal analysis (DiMitri et al., 2018). Further, with data collected over a longer timeframe in multiple school classrooms, SWPBS could identify patterns between student groups, curriculum activities, and primary school interiors with precise and objective sensor readings.

Some of these opportunities should rightly, and perhaps as obviously, be considered dangerous if not handled with care, as pointed out by researchers such as Prinsloo and Slade (2017, 2013). However, handling with care to avoid error does not mean advocacy of approaches primarily motivated by error-avoidance (Markham, 2018). Rather than error-avoidance, parts of this study’s outlined and applied methodology may be suitable as a developmental basis for further research in the learning sciences. This includes a contextual emphasis that considers actionable knowledge on a case-by-case basis (Cukurova et al., 2020).

For further teacher aid, the type of stress reading analysis provided in this study has clear potential to be integrated into a visual dashboard that might provide a useful service for teachers that monitor student group stress levels (de Arriba-Pérez et al., 2017; Noroozi et al., 2019). Such services may include intuitive ways to improve not only the understanding of students’ well-being or academic performance but also other aspects seen in the findings, such as social group formations.

Theoretical Implications

For the scope of this study, a case-by-case basis may relate to the importance of emphasising teachers’ ability to adjust curricular activity and corresponding sequential session structure as they see fit, as also indicated in previous PBIS-related research (Algozzine et al., 2021; Beamish & Bryer, 2019; Ferri & Ashby, 2017). Analysis with aims of larger scope towards generalised stress level optimisations for student learning in primary education curricular activities also needs a richer set of contexts, such as considerations to the whole day of students’ school experience, including differences between morning and afternoon resource expenditure. This study assumed the desirability of promoting an active sympathetic nervous system, as students with SEN often need support for this ability during curricular activities. However, it is important when considering comprehensive eco-psychological stressors for an applied behavioural analysis that the parasympathetic nervous system may be even more necessary to emphasise in order to enable proper recovery and rest (Ernst, 2014). Optimal PBIS actions also need to rely on the different tiers in the SWPBS tier system to adjust intervention intensity (Algozzine et al., 2021; Beamish & Bryer, 2019; Tutton, 2019). Gamified elements in curricular activities may aid in this regard, such as the Kahoot! implementations seen in this study, by generating rich contextual data (Palmquist et al., 2021).

Limitations

As discussed above, the biobehavioural data analysed in this study have ethical risks. Dire consequences for students could unveil if contextual emphasis and regulatory guidelines of GDPR or similar to GDPR are ignored, such as forceful surveillance of individual biobehavioural data as part of learning outcome assessment practices without relation to the well-being of students (Cukurova et al., 2020). To be absolutely clear, this article does not at this stage advocate for any other use of the presented findings than outlined in the results section. Ideally, sensitive data should be deleted as soon as its purpose is satisfied, and algorithmic analysis should only meet the eyes of a few with professional responsibilities. Reflection should be made on how education can provide special educational solutions for learning alongside technological developments that are ‘adaptive, flexible, and future-ready’ (Basham et al., 2020, p. 44). This includes discussing what educational benefits might outweigh the resources required to manage the risks of adopting technological innovations (Buckingham Shum & Luckin, 2019). While outlining further extensive ethical considerations is suitable, it is outside the scope of this article.

One key limitation of this study may relate to the validity of the ANS assumptions. For example, raising the sensor quality of individual student readings may require heavier clinical equipment and other methodological considerations. While the assumptions may be debated, heavier clinical equipment, on the other hand presents a challenging trade-off in interference with the naturalistic educational setting, as attachment of electrodes in electrocardiography methods have invasive elements such as movement restriction, skin irritations or skin injuries (Hogan & Baucom, 2016; Hoog Antink et al., 2021). However, this trade-off limits other uses of findings that may stumble upon risks of reliability, which scientific research that utilises less visually intuitive but more extensive statistical analysis could alleviate, such as spectral analysis (Akselrod et al., 1981; Hoog Antink et al., 2021), with synthesised modelling or multivariate computations across student group physiology correspondence and pairing, or curricular activities, or classroom interiors. These considerations could further combine perceptions from students’ self-reports, absent in this study and called out for in the literature, e.g. (Larmuseau et al., 2020).

Further Research

The findings and its PPG smartband methodology outline objective ANS stress measurements of a primary education student group with intuitive MMLA visualisations for actionable PBIS curriculum activity adjustments. One especially notable finding in this study was observable and consistent intensity spikes during activity shifts, which may have the potential for applications in research relating to students’ executive function (Bishara & Kaplan, 2022). Overall, the data has clear implications for future development, such as the proposed PBIS LA dashboard for teachers in a classroom, with innovative SWPBS implementations. The proposed dashboard relates to the, according to Cusumano and Preston (2021), needed aid for contextually identifying exactly when to execute PBIS actions.

While the presented results are promising, the proposed smartband methodology for primary education classrooms is still in its infancy and, therefore, not yet suitable to implement at scale for teachers. It is therefore necessary that further research focus on a bigger scope that includes environmental stressors, for example, to emphasise scalability. Additionally, ANS assumptions related to the behavioural mode variables may require further investigation for the scalability of the findings. However, if systematic challenges (most notably, ethical and validity details) are handled, the findings of this article serve as an empirical basis for viable cost-effective solutions with advantages to a greater understanding of student learning. This understanding would benefit the lives of students with SEN through the continued hard work and iterated implementations of teachers conducting PBIS, practitioners of SWPBS, and researchers in the learning sciences.

Disclosure Statement

No potential conflict of interest was reported by the author(s).